IEEE Pervasive Computing

______http://www.computer.org SOFTWARE-BASED EVENT DETECTION, P. 95 IEEE PRESIDENT-ELECT CANDIDATES, P. 79 CS ELECTION CANDIDATES, P. 71 o aiainisrcin laecikhere click please instructions navigation For o aiainisrcin laecikhere click please instructions navigation For A Computer Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page BEF MaGS

Running in Circles Looking for a Great Computer Job or Hire?

Make the Connection - IEEE Computer Society Jobs is the best niche > Software Engineer > employment source for computer science and engineering jobs, with Member of Technical Staff > Computer Scientist hundreds of jobs viewed by thousands of the finest scientists each > Dean/Professor/Instructor month - in Computer magazine and/or online! > Postdoctoral Researcher > Design Engineer > Consultant

http://www.computer.org/jobs

IEEE Computer Society Jobs is part of the Physics Today Career Network, a niche job board network for the physical sciences and engineering disciplines. Jobs and resumes are shared with four partner job boards - Physics Today Jobs and the American Association of Physics Teachers (AAPT), American Physical Society (APS), and AVS: Science and Technology of Materials, Interfaces, and Processing Career Centers.

Editor in Chief Associate Editor in Chief, Computing Practices 2010 IEEE Computer Carl K. Chang Research Features Rohit Kapur Society President Iowa State University Kathleen Swigger [email protected] James D. Isaak [email protected] University of North Texas [email protected] [email protected] Perspectives Associate Editor Bob Colwell in Chief Associate Editor in Chief, [email protected] Sumi Helal Special Issues University of Florida Bill N. Schilit Web Editor [email protected] Google Ron Vetter [email protected] [email protected]

Area Editors Column Editors Software Technologies CS Publications Board Computer Architectures AI Redux Mike Hinchey David A. Grier (chair), David Steven K. Reinhardt Naren Ramakrishnan Lero—the Irish Software Bader, Angela R. Burgess, Jean- AMD Virginia Tech Engineering Research Centre Luc Gaudiot, Phillip Laplante, Databases and Information Education Web Technologies Dejan Milojičić, Linda I. Shafer, Retrieval Ann E.K. Sobel Simon S.Y. Shim Dorée Duncan Seligmann, Don Erich Neuhold Miami University San Jose State University Shafer, Steve Tanimoto, and University of Vienna Embedded Computing Roy Want Distributed Systems Tom Conte Advisory Panel Jean Bacon Georgia Tech Thomas Cain CS Magazine University of Cambridge Green IT University of Pittsburgh Operations Committee Graphics and Multimedia Kirk W. Cameron Doris L. Carver Dorée Duncan Seligmann Oliver Bimber Virginia Tech Louisiana State University (chair), David Albonesi, Johannes Kepler University Linz Industry Perspective Ralph Cavin Isabel Beichl, Carl Chang, High-Performance Computing Sumi Helal Semiconductor Research Corp. Krish Chakrabarty, Vladimir Getov University of Florida Dan Cooke Nigel Davies, Fred Douglis, University of Westminster IT Systems Perspectives Texas Tech University Hakan Erdogmus, Lars Jentsch, Information and Richard G. Mathieu Ron Hoelzeman Carl E. Landwehr, Simon Liu, Data Management James Madison University University of Pittsburgh Dejan Milojičić, John Smith, Naren Ramakrishnan Invisible Computing Naren Ramakrishnan Gabriel Taubin, Fei-Yue Wang, Virginia Tech Albrecht Schmidt Virginia Tech and Jeffrey R. Yost Multimedia University of Duisburg-Essen Ron Vetter Savitha Srinivasan The Known World University of North Carolina, IBM Almaden Research Center David A. Grier Wilmington Networking George Washington University Alf Weaver Ahmed Helmy The Profession University of Virginia University of Florida Neville Holmes Software University of Tasmania Robert B. France Security Colorado State University Jeffrey M. Voas David M. Weiss NIST Iowa State University

Editorial Staff Contributing Editors Design and Production Administrative Senior Business Scott Hamilton Lee Garber Larry Bauer Staff Development Manager Senior Acquisitions Editor Bob Ward Design Products and Sandy Brown [email protected] Nancy Talbert Olga D’Astoli Services Director Senior Advertising Judith Prow Cover Design Evan Butterfield Coordinator Managing Editor Kate Wojogbe Senior Manager, Marian Anderson [email protected] Editorial Services Chris Nelson Lars Jentsch Senior Editor Manager, James Sanders Editorial Services Senior Editor Jennifer Stout

Circulation: Computer (ISSN 0018-9162) is published monthly by the IEEE Computer Society. IEEE Headquarters, Three Park Avenue, 17th Floor, New York, NY 10016-5997; IEEE Computer Society Publications Office, 10662 Los Vaqueros Circle, PO Box 3014, Los Alamitos, CA 90720-1314; voice +1 714 821 8380; fax +1 714 821 4010; IEEE Computer Society Headquarters, 2001 L Street NW, Suite 700, Washington, DC 20036. IEEE Computer Society membership includes $19 for a subscription to Computer magazine. Nonmember subscription rate available upon request. Single-copy prices: members $20.00; nonmembers $99.00. Postmaster: Send undelivered copies and address changes to Computer, IEEE Membership Processing Dept., 445 Hoes Lane, Piscataway, NJ 08855. Periodicals Postage Paid at New York, New York, and at additional mailing offices. Canadian GST #125634188. Canada Post Corporation (Canadian distribution) publications mail agreement number 40013885. Return undeliverable Canadian addresses to PO Box 122, Niagara Falls, ON L2E 6S8 Canada. Printed in USA. Editorial: Unless otherwise stated, bylined articles, as well as product and service descriptions, reflect the author’s or firm’s opinion. Inclusion in Computer does not necessarily constitute endorsement by the IEEE or the Computer Society. All submissions are subject to editing for style, clarity, and space.

AUGUST 2010 1

CONTENTS For more information on computing topics, visit the Computer Society Digital Library atwww.computer.org/csdl. Library Digital Society Computer the visit topics, computing on information more For Services Hold? Will Limits What COVER FEATURES PERSPECTIVES 20 27 37 rvosPg otns|Zo n|Zo u rn oe erhIse|Nx Page Next | Issue Search | Cover Front | out Zoom | in Zoom | Contents | Page Previous rvosPg otns|Zo n|Zo u rn oe erhIse|Nx Page Next | Issue Search | Cover Front | out Zoom | in Zoom | Contents | Page Previous challenges. privacy and security several present also they Web, Social but inthe data user enrich to and connectincreaseSocial-network services access Mohamed Shehab, and Ravi Sandhu P. Ko, Gorrell Nam Moo Cheek, Social-Networks Connect computing. multimedia social called area, research content on the Internet has led to a new multimedia ofsocial growth explosive The Tiejun Huang, and Noshir Contractor Yonghong Tian, Jaideep Srivastava, Social Multimedia Computing components. unreliable from systems reliable creating as such paradigm, new anentirely seek also must industry the design. Eventually efficient inenergy- abreakthrough require will pattern Moore’slaw, followed that continuing but has technology semiconductor years, 40 For Izydorczyk Michael and Izydorczyk Jacek Scaling: Microprocessor 27 http://computer.org/computer Expansion of Knowledge Online Search Engine: Crowd-Powered on an Intercontinental Grid Intercontinental onan Computing with Ibis and its uniquely rich online/offline rich interactions. uniquely its and tremendousin recent its growth examines years inChina thatoriginated function asearch of study empirical comprehensive first This RESEARCH FEATURES I on the grid, and the limits ofMoore’slaw. limits the and grid, the on real-world distributed comput uniquely rich online/offline interactions. We also look at its inChina—and thatoriginated function a search ofknowledge— expansion ofcrowd-powered study cal Web; anempiri- Social and inthe data user enrich to and access thatincrease services connect social-networking new technology; multimedia and science social the ABOUT THIS ISSUE THIS ABOUT 54 63 45 a new multidisciplinary research area that bridges areathatbridges research multidisciplinary a new n this issue, we at look so A Study of the Human Flesh Flesh A Study of Human the Simulating the Universe Real-World Distributed Yanqing Gao, Hui Wang, and Guanpi Lai Guanpi and Wang, Hui Yanqing Gao, Hendler, Qingpeng Zhang, Zhuo Feng, A. James Zeng, Fei-Yue Daniel Wang, world apart. apart. world a half supercomputers two on processing concurrent totry team ofresearchers aninternational led universe ofthe a sector ofsimulating requirements computational The Grosso Paola and Harfst, Stefan Kei Hiraki, McMillan, Stephen Laat, de Cees Makino, Junichiro Nitadori, Keigo Groen, Derek Tomoaki Ishiyama, Zwart, Portegies Simon environments. heterogeneous faulty, and dynamic, even for applications, distributed compute-intensive of Ibis prog allows easy and Kees Verstoep Jacobs, J.H. Ceriel Frank J.Seinstra, Thilo Kees Kielmann, van Reeuwijk, Wrzesi Palmer, Gosia Nick Kessel, van n Niels Drost,Nieuwpoort, Roelof Kemp, Timo V. Rob van Maassen, Jason Bal, E. Henri cial multimedia computing, computing, cial multimedia ramming and deployment ing, scientific computing ing, scientific ´ska, ´ska, A A B B E E F F M M a a G G S S A Computer Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page BEF MaGS

Flagship Publication of the IEEE IEEE Computer Society: http://computer.org Computer: http://computer.org/computer Computer Society [email protected] IEEE Computer Society Publications Office: +1 714 821 8380 August 2010, Volume 43, Number 8

6 The Known World Mental Discipline David Alan Grier 9 32 & 16 Years Ago Computer, August 1978 and 1994 Neville Holmes NEWS 98

11 Technology News 98 Education Researchers Fight to Keep Implanted Computer Science: Is It Really the Scientific Medical Devices Safe from Hackers Foundation for Software Engineering? Neal Leavitt Stephen T. Frezza 15 News Briefs 102 Industry Perspective Linda Dailey Paulson IBM’s University Programs MEMBERSHIP NEWS James C. Spohrer 71 IEEE Computer Society 106 Invisible Computing Election The Hardware Is Not a Given Steve Hodges and Nicolas Villar 79 IEEE President-Elect Candidates 110 Software Technologies 88 Call and Calendar Data-Intensive System Evolution Anthony Cleve, Tom Mens, and Jean-Luc Hainaut COLUMNS 116 The Profession 91 Security The Coming Robot Crime Wave From Chaos to Collective Defense Noel Sharkey, Marc Goodman, and Nick Ross James Bret Michael, Eneken Tikk, Peter Wahlgren, and Thomas C. Wingfield DEPARTMENTS 95 AI Redux 4 Elsewhere in the CS Computer Society Information Preventing Future Oil Spills with Software- 19 Based Event Detection 84 Career Opportunities S.S. Iyengar, Supratik Mukhopadhyay, 87 Advertiser/Product Index Christopher Steinmuller, and Xin Li 90 Bookshelf

Reuse Rights and Reprint Permissions: Educational or personal use of this material is permitted without fee, provided such use: 1) is not made for profit; 2) includes this notice and a full citation to the original work on the first page of the copy; and 3) does not imply IEEE endorsement of any third-party products or services. Authors and their companies are permitted to post their IEEE-copyrighted material on their own Web servers without permission, provided that the IEEE copyright notice and a full citation to the original work appear on the first screen of the posted copy. Permission to reprint/republish this material for commercial, advertising, or promotional purposes or for creating new collective works for resale or redistribution must be obtained from the IEEE by writing to the IEEE Intellectual Property Rights Office, 445 Hoes Lane, Piscataway, NJ 08854-4141 or [email protected]. Copyright © 2010 IEEE. All rights reserved. Abstracting and Library Use: Abstracting is permitted with credit to the source. Libraries are permitted to photocopy for private use of patrons, provided the per-copy fee indicated in the code at the bottom of the first page is paid through the Copyright Clearance Center, 222 Rosewood Drive, Danvers, MA 01923. IEEE prohibits discrimination, harassment, and bullying. For more information, visit www.ieee.org/ web/aboutus/whatis/policies/p9-26.html.______

ELSEWHERE IN THE CS

Computer Highlights Society Magazines

he IEEE Computer Society offers a lineup of 13 peer-reviewed technical magazines that cover cutting-edge topics in computing Organizations that have typically attempted to effect T including scientific applications, design and strategic change through initiatives aimed at people, test, security, Internet computing, machine intelligence, processes, or technology are discovering that more mean- digital graphics, and computer history. Select articles from ingful and lasting change is achieved through a focus on recent issues of Computer Society magazines are high- culture, organization, and governance. IT, as the “enabler,” lighted below. has a unique view across and through the enterprise and can provide the best input to executive leadership on opportunities for change. CIO Corner contributor Tom Costello of UpStreme looks at Two well-known authors in software engineering, Barry the role of IT in governance in “COG vs. PPT: Models for Orga- Boehm and Kent Beck, offer their complementary perspec- nizational Change” in the July/August 2010 issue of ITPro. tives on software evolution in “Perspectives on Software Evolution: Its Changing Nature, and Its Inevitability’ in the July/August 2010 issue of Software. Boehm argues that the time of “one software evolu- The world as we know it has collapsed into custom-sized tion process fits all” is over and provides guidelines on computer screens with Mercator-projected maps, satellite selecting the most appropriate evolution-friendly process images, aerial photos, and graphics showing us where every- under various circumstances. Beck acknowledges the dif- thing is, what every location looks like, and why we would ficulty of evolving a software design. He reminds us of the want to go there. Bing Maps combines route information with factors that software developers and managers must con- aerial and street-level photography to provide a seamless, sider when evolving systems (such as cost, time, and risk), interactive canvas for visualizing location information. Chris including the need to keep a system operational during its Pendleton of Microsoft explores Bing Maps in “The World actual evolution, which is particularly relevant for some According to Bing” in the July/August 2010 issue of CG&A. critical applications.

The July/August 2010 issue of CiSE is based on work The 16th Asian Games in Guangzhou, China, in presented at the US National Science Foundation workshop November, will involve 42 sports and more than 100,000 Path to Petascale: Adapting Geo/Chem/Astro Applications people—the largest such games ever. A parallel event for for Accelerators and Accelerator Clusters, held at the US disabled Asian athletes will start two weeks later. National Center for Supercomputing Applications in early Transportation is a major challenge facing the 2010 2009. The workshop was designed to raise awareness about Asian Games. Since the events will use 58 existing the emergence of accelerator-based high-performance facilities and 12 new stadiums located widely across computing resources among computational scientists from the metropolitan area, safe, effective transportation the geosciences, computational chemistry, and astronomy management will be essential. Read about Guangzhou’s and astrophysics communities and to help them get started intelligent transportation systems approach, designed to in using these resources. make the games a success, in “Parallel Traffic Manage- “High-Performance Computing with Accelerators” in ment for the 2010 Asian Games,” in the May/June 2010 the July/August 2010 issue of CiSE looks at recent develop- issue of IS. ments in the field.

Tracking organizations such as US-CERT report a con- Micro’s May/June 2010 issue captures two debates from tinuing rise in security vulnerabilities in software. But 2009 Computer Architecture Research Directions, a work- not all discovered vulnerabilities are equal—some could shop at the 36th International Symposium on Computer cause much more damage to organizations and individuals Architectures. In “The Future of Architectural Simulation,” than others. In the inevitable absence of infinite resources, Intel’s Joel Emer and Microsoft Research’s Doug Burger software development teams must prioritize security for- spearhead a debate on architectural simulation’s future tification efforts to prevent the most damaging attacks. in computer research. In “Programming Multicores: Do Protection Poker is a collaborative means of guiding this pri- Application Programmers Need to Write Explicitly Parallel oritization. “Protection Poker: The New Software Security Programs?,” Princeton University’s David August debates ‘Game’” in the May/June 2010 issue of S&P is a case study of Keshav Pingali from the University of Texas at Austin. how a Red Hat IT software maintenance team demonstrates Protection Poker’s potential for improving software security practices and team software security knowledge. In “Real-Time Environmental Monitoring and Notifi- cation for Public Safety” in the April–June 2010 issue of MultiMedia, Kean University researchers describe a low- Mobile devices and social media have considerable cost system that combines environmental monitoring and potential for facilitating learning, from both the indi- Web-based real-time data reporting. The Wireless Inte- vidual-skills and socialization perspectives. However, grated Network Sensors database application provides acceptable-use policies have limited the use of mobile access to and visualization support for both real-time and devices on school campuses as a response to the risks archived data. The WiNS prototype samples environmental schools face in dealing with disruptive or harmful speech. voltage, humidity percentage, temperature, atmospheric In “Acceptable Use of Technology in Schools: Risks, pressure, barometric pressure, and light. Learning func- Policies, and Promises” in the July-September 2010 issue tions are part of further development under way. of PvC, authors Meg Cramer and Gillian Hayes of the Uni- versity of California, Irvine, ask why mobile devices and reduced 70%

social media applications are much less pervasive in the reduced 70% classroom than in other parts of youth life. Educators, In a special July/August issue of D&T, “Emerging Inter- researchers, and designers may need to work together to connect Technologies for Gigascale Integration” highlights increase understanding of youth’s experience with perva- recent investigations of revolutionary interconnect para- sive computing technologies and provide greater access to digms for future SoCs and other computing platforms. these systems and applications in a formal school context. Guest editors Partha Pratim Pande of Washington State University and Sriram Vangal of Intel have selected five articles representing a wide range of emerging intercon- nects—from carbon nanotubes to optical, RF, and on-chip Information and communication technology (ICT) is a wireless communications. crucial driver of economic growth. Authors Stephen Ezell and Scott Andes of the Information Technology and Inno- vation Foundation look at current and future investments in ICT around the world in “ICT R&D Policies: An Interna- In “Unraveling Algol: US, Europe, and the Creation of a tional Perspective” in the July/August 2010 issue of IC. Programming Language” in the latest issue of Annals, David While the US still performs the most ICT R&D in the Nofre from the University of Amsterdam revisits the early world, competition has intensified as US ICT R&D invest- development of Algol (1955–1960). He takes issue with the ments as a percentage of GDP have fallen noticeably and American-European and academic-commercial polarities have been surpassed by competitors in the past decade. that usually characterize the demise of a universal pro- A Web extra sidebar explains ICT R&D intensity, which gramming language and concludes, “If the history of Algol measures a country’s performance based on its R&D calls for a dichotomous interpretation, it was a matter of investment compared with the size of its economy. uniformity versus diversity.”

Editor: Bob Ward, Computer;[email protected]

AUGUST 2010 5

THE KNOWN WORLD

Mental Discipline David Alan Grier, George Washington University

The practices of engineering and computer science are influenced by the same forces that shape manual labor and office work. Occa- sionally, it’s useful to reassess our skills and question the value of our training.

t was a barrier to entry. Even to join our party. He laughed at our the inquisitive, a comfort of man, and in my neighborhood, which jokes, agreed with our comments, a protection to women.” occasionally leans toward a and attempted to add his opinions dictatorship of the proletariat, to our discussions. Just as the dinner THE END OF MATHEMATICS? I the political philosophy of the came to an end, he moved a chair to Mathematics may not be the most Running Dog Café can discourage our table and sat next to a woman in universal way of protecting individu- patrons. The Running Dog was run our party named Carey, who seemed als from unwanted attention, but it as a Marxist collective and promoted to be the object of his attentions. has long served as a barrier to dis- its ideology at the front door. “If you He chose a bad moment to tinguish one group from another. find any word, item, or action to be make his move. At that instant, the First-year engineering students bond racist, sexist, ageist, culturalist, or comrade-waiter delivered our bills, over their struggles with calculus, in any other way designed to divide handing one to each member of the and economists wear their statistical the laboring classes,” read the sign table. When we looked at the pieces knowledge as a badge of honor. The over the cash register, “please notify of paper in our hand, we realized that practice is as least as old as Napoleon, a member of the wait staff, who will each of us had received the wrong bill. who required all his artillery officers raise the issue at the next meeting of Seeing his opportunity, the unfor- to learn mathematics, even those who the Central Committee.” tunate young man asked, “Hey, what might not be expected to use it. “This The stern warning camouflaged a is the probability of that happening?” particular shared body of knowl- lovely restaurant. The food was good, We all paused and then immedi- edge,” explained the historian Ken the atmosphere congenial, and the ately began searching for pens and Alder, “distinguished the artillerists, staff pleasant. It was a great place to pieces of scratch paper. With just a as a body, from the general officers.” hold a lunch meeting, provided you few minutes of work, we determined Yet, in the past month, I’ve been were willing to let other customers that there were three obvious ways of part of three conversations in which eavesdrop on your conversation. calculating the probability. You could the participants honestly discussed Apparently, many of the regulars enumerate the sample space, employ removing mathematics from the subscribed to the philosophy, more a combinatorial argument, or do a engineering and computer science associated with USA Today than the clever conditional analysis. curriculum. A dinner with friends. Daily Worker, that your business is my Carey led the discussion of the A discussion with an engineering business. problem, as I recall. As she finished, dean. The argument to remove math- It was often difficult to keep prying she looked at the empty chair next ematics is based on the claim that eyes and open ears away from your to her and realized that the trouble- most mathematical analysis is now conversations. At one of the last din- some young man had disappeared. handled by computer programs. ners I had at the Running Dog, I was He had apparently retreated during “We would never do mathemat- subject to a running commentary our computations. “Such is the power ics analysis by hand,” the argument from a young man who was trying of mathematics,” she said. “An aid to usually begins. “We would always

use a mathematical analysis pack- ence. When teachers claim that their zie. “Increasing black-boxing may, age, a structural modeling program, subject merely disciplines the mind, indeed, be a passable definition of or statistical software. We would they’re admitting that their ideas are modernization.” never trust any calculation done by no longer relevant for most students. a person.” Still, our relationship to math- AID TO THE INQUISITIVE As you can imagine, such sug- ematics has clearly been shifting While it’s possible to trace the idea gestions are met by an explosion over the past two decades because of of the black-box to the origin of the of emotion that can’t admit to such forces that computer scientists have machine or the invention of inter- a change. This emotion is usually unleashed through a process known changeable parts, we usually identify followed by three objections, most as black-boxing. the concept with the electrical engi- forcefully made, for the retention of neer Wilhem Cauer (1900-1945), who mathematics. First is the claim that BLACKfiBOXES developed black-box techniques to engineers need to be able to double- Black-boxing is an activity that simplify the work of designing elec- check calculations. The second objec- hides expertise. It takes a body of tronic circuits. Cauer’s biographers tion states that it’s useful to estimate knowledge, incorporates those ideas note that during the 1920s, “engineer- the order of magnitude for any calcu- in a machine, and then hides the inner ing problems were typically solved lation. The final, and most persistent, operation of that machine behind an in an empirical fashion by analyzing is the idea that mathematics pro- opaque façade. To use the knowledge, specific circuits instead of looking vides a mental discipline for young we operate the black-box through a at classes or families of circuits.” engineers. simplified set of controls. Cauer recognized that certain elec- Mathematics will likely be tronic functions could be isolated in retained but not for any of these We’ve long observed a simple device that had only one or three reasons. The first two are spe- two or four connections to the rest cious, and the last is an admission how black-boxing of the circuit. Though he never used of defeat. If you don’t accept hand has reduced the the term, the isolated device would be calculation as part of the original demand for expensive known as the black-box. analysis, you aren’t going to accept mathematical skills in In Cauer’s day, black-boxes would it as a validation machine-computed the commercial world. be filters, circuits that would block results. In many cases, it’s difficult electronic signals in a certain range to compute even a meaningful of frequencies. To build these fil- estimate of a final answer, as the Programming languages are clas- ters, Cauer created a unified theory analysis can involve nonlinear dif- sic examples of black-boxes. They of filter design. Engineers could use ferential equations or large matrices. incorporate knowledge about com- this theory to design filters without The third defense of mathematical puter architecture: memory, registers, worrying about the details of the education, the claim that it disciplines machine codes, and all the other construction. They could specify the the mind, is the last defense of a dying aspects of actual physical devices. filter and then grind through a series field. The proponents of Greek, Latin, They allow individuals who know of equations to produce the design. music theory, drafting, public rhet- nothing about the inner workings “All known wave filters” of a certain oric, and a host of other topics that of a computer to utilize knowledge class “are contained in the filters were once required subjects of study about computing logic and produce of the new theory,” he explained. have tried to retain a privileged place useful output. This “new theory has been carried in the curriculum for their subjects by Application packages such as through practically to find out the arguing that Ovid and Virgil and the spreadsheets, word processors, and most economical filter for any prac- sonata form were a good discipline games are other examples of black- tical purpose.” for students. None of them succeeded. boxes, as they encapsulate knowledge The black-boxing of filters is often In American education, educa- about programming. Indeed, many treated as a technical achievement, tional historian Laurence Veysey if not most engineered products can but it’s really a step in the division of argued the “idea of mental discipline be considered black-boxes—ther- labor, of engineering labor. “When contained inherent weaknesses as mostats, automobiles, mp3 players, a good circuit designer lays out a a conception,” as it emphasizes the and inertial guidance systems. All of new circuit,” explained a Bell Labo- esoteric over the practical, subjects them can be viewed as systems that ratories engineer, “he isolates those that are distant from daily experi- hide knowledge. “Black-boxes litter functions that are more or less self ence over those that help students the societies of high modernity,” contained” so it’s easy “to partition master physical and social experi- notes sociologist Donald Macken- the problem so that several people

AUGUST 2010 7

THE KNOWN WORLD

can work on the different parts.” tions ago, sales clerks needed to know used by Napoleon and, I suppose, by This partition allows for a division of enough mathematics to be able to Carey. Mathematics sets us apart—it’s labor. It lets a senior engineer assign sum a list of prices, to calculate taxes, a barrier against the unwashed. Math- simple tasks to less experienced to make change. Step by step, cash ematics makes us special. designers. It also allows any designer registers and point-of-sale terminals Not that long ago, I attended an to make use of expertise that a design black-boxed that skill. Now most sales event where I found myself in the team might not possess. If done well, clerks need only know how to operate company of a former high-level gov- the partition should allow the team the box, how to aim a laser at a Univer- ernment official. This gentleman was to replace parts of the circuit with sal Product Code, and how to record very proud of his own accomplish- commercial modules. “Not only is a credit card. A few senior clerks will ment and was quite willing to divide this cheaper because of saving on have advanced training on the black- the social distance between the two design time,” he continues, “but it boxes of the sales terminals and will of us by reciting his own resume. He may provide a better module than know how to control their operation, had been ambassador to a vital ally, he could have produced. The func- but their number will be few. deputy ambassador to a dangerous tion of this module may not be the As in the world of commerce, black- enemy, undersecretary for important designer’s specialty, whereas the boxing divides labor in the world of stuff, and special assistant for pow- designer of an off-the-shelf module engineering and computer science. erful people. He was with Reagan in is likely to have been a specialist in Tools such as Mathematica, Maple, Helsinki, Nixon in China, and Ken- that subject area.” and AutoCad split the population of nedy during the Cuban Missile Crisis. Cauer’s black-boxes split the engi- technical laborers into a large group In 40 seconds and five sentences, he neering community into two groups: that can take its mathematical skills let me know that he was important specialists and generalists. The funda- from software and a smaller cohort and that I was not. mental benefit of such a division was that can contribute their knowledge “What have you done?” he asked identified by Charles Babbage in 1832. to those black-boxes. If we’re to adjust with a grin that was friendly in the By “dividing the work to be executed the skills that we require of engineers sense that it suggested that I had into different processes, each requir- and computer scientists, it will have already lost the fight. “Not much,” I ing different degrees of skill or of to be done in relation to those tools admitted. “I’m a computer scientist force,” he observed, an employer “can rather than follow the argument that and have a PhD in mathematical purchase exactly that precise quantity the rising generation needs to have statistics.” of both which is necessary for each its mind disciplined or that it needs He didn’t quite slink away, like the process.” Employers no longer need the same kind of contact with math- young man who had tried to win Car- to pay for expensive skills when skills ematical symbols that we had, or our ey’s attention, but he quickly found are not needed. Those skills can be parents had, or, in my case, my grand- an excuse to leave my presence. the understanding of circuit design, mother had. He departed, muttering something the mastery of difficult computer about never being very good at math. languages, or even the mastery of COMFORT TO MAN Mathematics may have lost its ability mathematics. Of course, if we don’t want to face to help us with common problems, We’ve long observed how black- the suggestion that our software has it may have lost some of its value to boxing has reduced the demand for removed a fundamental set of skills engineers, and it may not be able to expensive mathematical skills in the from a portfolio of engineers, we discipline our mind, but it makes us commercial world. Just three genera- can always retreat to the reasoning special and occasionally can be used to discipline the minds of others.

Not only does David Alan Grier have a degree in the mathematical sciences, technology news to you. but so did his grandmother (Blanch O’Kane, University of Michigan class of Let us bring ’21). He writes more about these subjects on the blog “The Known World,” www.computer.org/theknownworld. http://computingnow.computer.

Subscribe to our daily newsfeed Selected CS articles and columns are available for free at http:// org ___ ComputingNow.computer.org.______

8 COMPUTER

32 & 16 YEARS AGO

AUGUST 1978

SPECIAL MESSAGE (p. 3) “One of the primary goals of the MUSIC COM- IEEE Computer Society is the dissemination of informa- POSITION (p. tion. To this end, over two-thirds of the Computer Society 40) “A useful budget is allocated for publications. In 1977, we published place to begin in the neighborhood of 20,000 pages—by far the largest examining the output among all IEEE societies and groups.” field of computer music composition THE HUMANITIES (p. 6) “Computing in the humanities is to focus on three is a field dominated by amateurs, in the best sense of conventional and somewhat this word. Nothing forces a critic to put his texts on a arbitrary subdivisions: sound synthesis, compositional computer; no composer is compelled to seek the aid of structure, and composer-machine communication. . . . By a machine; even the programmers employed on this kind specifying sound synthesis and compositional structure, of project are likely to be there by inclination rather than we follow the traditional division of music into sound by accident. Economic motives are also largely absent: and structure, by which we mean the acoustic repertoire, in general, nobody makes or saves any money by using and the set of relationships between elements of this rep- computers for such applications, and only occasionally ertoire. The arbitrary sound-structure distinction, made can the machine save time.” mainly for convenience of representation, should not prevent us from considering methods of musical organization HISTORY (p. 8) “What currently draws students of history, where sound and structure cannot be easily separated, as in ever growing numbers, to the use of the computer? The is often the case in electronic music.” attractiveness of the computer seems to have two quite diverse roots. The computer promises aid to the empiri- PACKAGING (p. 66) “. . . Multiple-chip packages have cal researcher who seeks to master, organize, and exploit appeared in recent commercial products, namely the large documentary funds. And it also supports, but in a 8K-bit RAM in the IBM 5100 and the 16-bit Teledyne Sys- much different fashion, the aspirations of social theorists, tems TDY52 microcomputer, yet most manufacturers who seek to construct, on the basis of historical data, appear to be shying away from multichip packages for mathematical or logical models of past human behavior. the time being. Current efforts seem to focus on obtain- While the work of these two groups is, or ought to be, ulti- ing the maximum density and area possible with existing mately convergent, nevertheless each of them produces technologies. Future packaging, especially for memory, its own special style of quantitative history. These two will undoubtedly use multichip techniques after single- differing styles of current work in quantitative history chip limits in density and area have been reached. The merit further examination.” realm of a future computer architect may well include structuring complex package topology.” CHOREOGRAPHY (p. 19) “My first experiments with computer-generated dance produced sequences that were TOMOGRAPHY (p. 79) “Union Carbide’s radionuclide body pleasing to both dancers and viewers, but they provided function imager provides 128 × 128 pixel computed- for too little human participation, while running up com- tomographic reconstructions of slices of the body organ puter time charges beyond my means. Later experiments, under study by a nuclear physician. In practice a radio- as you will see, struck what I felt to be a better balance pharmaceutical specific to the organ to be scanned is between human and computer participation.” injected, and the patient, on the gantry bed, is moved through the imager. Scanning takes from two to five min- LITERATURE (p. 32) “It is not useful to complain about utes per slice.” the scarcity of funds for computer-aided research in the humanities. The American Council of Learned Societies SPEECH RECOGNITION (p. 85) “The Illinois Department dropped this as a stated category of grants a few years ago. of Administrative Services has implemented a telephone When money was forthcoming, some of the results were control system in which a state employee seeking to place well received, even in traditional circles. The series of con- a credit card phone call talks to a computer, which han- cordances produced at Cornell University, the archives of dles the transaction and records all pertinent credit data.” Latin and Greek texts gathered at Dartmouth and California (Irvine), the JEUDEMO text-analyzing system at Montreal, and the COCOA Concordance package from Atlas Computer Editor: Neville Holmes; [email protected]______Laboratory in Britain are obvious examples.”

32 & 16 YEARS AGO

AUGUST 1994

EDITOR’S FORECAST (p. 6) “Consumers ignore PowerPC, tainability models to 11 industrial software systems and Taligent, IBM, Motorola, Adobe, Novell, HP, and Apple, used the results for fact-finding and process-selection and instead follow Microsoft, Lotus, Compaq, Dell, and decisions. The results indicate that automated maintain- Intel as they lead the world down the x86/Windows 4.0 ability assessment can be used to support buy-versus-build path. In the year 2000, Intel brings out the PA-RISC chip. decisions, pre- and post-reengineering analysis, subcom- Customers loudly complain as Microsoft and Lotus take ponent quality analysis, test resource allocation, and the two years to port their applications into the RISC world. prediction and targeting of defect-prone subcomponents. While Bill Gates celebrates the purchase of Italy (for vaca- Further, the analyses can be conducted at various levels tions), Apple, IBM, Motorola, and their followers file for of granularity.” bankruptcy. Customers display bronzed Sun, DEC, and HP workstations in glass displays as a reminder of the SOFTWARE QUALITY (p. 50) “For software developers, good old days.” early estimates of software quality are crucial, particularly on large-scale projects such as the space shuttle ight INTELLIGENT AGENTS (p. 8) “‘Intelligent agents’ will do software. Projects benefit when software managers can for software in the late 1990s what graphical user inter- identify software that does not meet quality requirements faces did in the 1980s according to Ovum, a London-based early in the development process so that corrective action research firm. Ovum’s recent report, ‘Intelligent Agents: can be taken when costs are low. Unfortunately, direct the New Revolution in Software,’ predicts that the develop- measurements of software quality (for example, reliability ment tool market for intelligent agents will grow from $11 or time to failure) can only be collected late in a project. million this year to $66 million in 1996 and $653 million Thus, the need arises to collect and validate metrics early by 2000.” in a project and use them to predict and control software quality during design.” PARALLELISM (p. 13) “Although multiprocessors can provide high computational rates, using them efficiently FUNCTION POINTS (p. 66) “Since its publication in Octo- requires program partitioning and restructuring for par- ber 1979, use of the function point metric has exploded allel execution. Parallelizing compilers can restructure throughout the software community. By 1993, it had programs automatically to extract simple loop-level paral- become the most widely used metric in the history of soft- lelism, but most codes (including those used in scientific ware. The nonprofit International Function-Point Users applications) are not amenable to compiler-based paral- Group (IFPUG), incorporated in 1984, has been expanding lelization. They require the exploitation of unstructured at a rate of 50 percent per year. At the end of 1993, there task-level parallelism, which must be indicated by the were user groups in about 18 countries, and in each of programmer. We present the programming language these countries the function point metric was the most COOL (Concurrent Object-Oriented Language), which was widely used measure of software productivity and quality designed to exploit coarse-grained parallelism at the task data normalization.” level in shared-memory multiprocessors.” ACCOUNTABILITY (p. 104) “An electronically equipped DATAFLOW COMPUTING (p. 27) “Due to its simplicity bureaucrat constitutes a potentially profound threat to and elegance in describing parallelism and data depen- public safety. In his or her hands, EDI [Electronic Data dencies, the dataflow execution model has been the Interchange] may be more nefarious than some lunatic’s subject of many research efforts. Since the early 1970s, semiautomatic assault weapon at a World Cup match. a number of hardware prototypes have been built and The bureaucracy favors the EDI initiative for it fulfills evaluated, and different designs and compiling tech- the perpetual vision of concentrated power with reduced niques have been simulated. The experience gained from accountability. Civil and criminal penalties must be these efforts has led to progressive development in data- stiffened to deter malicious or cavalier misuse of EDI ow computing. However, a direct implementation of systems for procurement, audit, or operational purposes. computers based on the dataow model has been found Bureaucrats must be held personally accountable for all to be an arduous task.” fraudulent EDI procurement actions.”

SOFTWARE METRICS (p. 44) “The intent of this article is to demonstrate how automated software maintainability PDFs of the articles and departments of the August issues of analysis can be used to guide software-related decision Computer for 1978 and 1994 are available through the IEEE making. We have applied metrics-based software main- Computer Society’s website: www.computer.org/computer.

10 COMPUTER

TECHNOLOGY NEWS Researchers Fight to Keep Implanted Medical Devices Safe from Hackers

Neal Leavitt

Implantable medical devices have become increasingly popular, and a growing number are equipped with wireless communications technology to increase their usefulness. However, this could make the devices susceptible to hackers.

mplantable medical devices— a research scientist at the US Depart- However, the risk is growing, as is such as insulin pumps, cardiac ment of Energy’s Oak Ridge National the number of patients using IMDs pacemakers, and cardiac Laboratory (ORNL). in part because of the aging of the defibrillators—have become All this convenience may come population. increasingly popular since with unanticipated risks: the pos- “The time to prevent future attack beingI introduced about 50 years ago. sibility that hackers could break scenarios is now,” said Paul. In the US alone, 2.6 million people into IMDs’ communications and “Hacking a medical device—espe- rely on IMDs. either send harmful commands to cially an implantable one—can have An increasing number of today’s the devices or steal private patient serious consequences and therefore devices are equipped with wireless information. must be taken seriously,” said Ed technology enabling, for example, A team of researchers from Har- Moyle, a senior manager with health- remote checks by healthcare workers. vard University, the University of care consultancy CTG. “Patients often receive at-home, Massachusetts Amherst, and the Uni- bedside monitors that wirelessly versity of Washington demonstrated INSIDE THE IMD collect telemetry from implanted in 2008 that hackers could extract The first implantable cardiac pace- devices,” said University of Massa- patients’ private medical information maker—developed by Medtronic chusetts Amherst assistant professor and reprogram their devices using founder Earl Bakken—was released Kevin Fu, codirector of the Medical off-the-shelf radio and computer in 1958. Device Security Center. equipment. IMDs have advanced over the The monitors relay stored informa- Currently, said Dougherty, the risk years. For example, the devices are tion to a server, which then makes the of malicious or otherwise unauthor- now capable of two-way wireless distilled data available to clinicians, ized manipulation of an implantable communications. in some cases via Web browsers, device is very low. Many insulin pumps use low- added Wendy Dougherty, program “To our knowledge, there has power chips with small transceivers director for public relations for IMD never been a single reported incident that send data—such as blood glucose vendor Medtronic. outside of controlled lab experiments levels—to other system components “Many of these devices now in more than 30 years of device and then receive commands to, for communicate with PCs to upload telemetry use,” she noted. instance, pump more insulin. stored information and may soon Currently, IMDs’ benefits outweigh In October 2005, Zarlink Semicon- communicate with devices such as their risk, so patients should use them ductor introduced the first transceiver smartphones,” said Nathanael Paul, if prescribed, said Paul. module designed explicitly for linking

TECHNOLOGY NEWS

Component 1: IMDs sometimes use off-the-shelf Bidirectional technologies for communications and other functionality. According to CTG’s Moyle, these underlying technologies’ vulnerabilities could affect the devices. “A primary concern is [hackers] eavesdropping on the communication channel between the device and external control units,” he noted. “Anytime you have a wireless data connection, you raise the possibility of this, as well Component 2: Unidirectional (send) as possible spoofing attacks.” Such attacks would let hackers emulate a legitimate part of an IMD system and Medical device obtain or alter information. Attackers could also intercept and record commands and then replay Component 3: them. Unidirectional (receive) Source: Nathanael Paul, Oak Ridge National Laboratory Potential motivations for hacking IMDs include the desire to harm Figure 1. Many of today’s implantable medical device systems, such as insulin pumps, either a specific person or just some- are equipped with wireless technology. The IMD shown here could communicate with one in general. a bidirectional component, such as a remote controller; a send-only device, like a blood Said Paul, “A public official or glucose monitor; or a receive-only device, such as a PC that acquires patient status celebrity could be attacked. A student information. These many lines of communications make IMDs vulnerable to hackers may even wish to skip a test and issue who could either send harmful commands to the devices or steal private patient data. some commands to the teacher’s medical device.” implanted medical devices and base companies even develop their own Other motivations, he noted, could stations. processors for complex calculations. include hurting an IMD maker’s repu- IMDs work with various radio tech- Today’s IMDs don’t connect tation or gaining personal satisfaction nologies that operate over distances directly to the Internet, although from hacking. of several centimeters and transmit some wirelessly connect to a bedside Hackers, added Moyle, could target within designated industrial, scien- monitor that then connects to the people who they know wear IMDs tific, and medical frequency bands. Internet. or they could also launch attacks in Most devices work with propri- crowded areas or near medical facili- etary communications protocols. DEVICES AT RISK ties, hoping someone with a device However, a few devices support the Connecting an IMD to computers is nearby. ZigBee wireless standard and vendors and phones makes treatment and may release medical applications monitoring more convenient. How- 2008 IMD hacking using Bluetooth, via the technol- ever, this can also make the device demonstration ogy’s Health Device Profile, soon, susceptible to attacks already faced The Harvard University, Univer- said University of Massachusetts by computers and phones, as Figure sity of Massachusetts Amherst, and Amherst doctoral candidate Benja- 1 shows. University of Washington researchers min Ransford. In 2003 and 2009, the Slammer in the 2008 IMD-hacking demonstra- IMDs often work with software- and Conficker worms infected some tion used inexpensive, off-the-shelf defined radios so that a single device networked hospital systems respon- Linux PC and GNU radio software to could, for example, operate over mul- sible for monitoring heart patients, intercept and capture the short-range tiple frequencies. said ORNL’s Paul. signals that an implantable cardiac They also use various types of He noted that anyone can com- defibrillator sent to an authorized processors, including those that run municate with an IMD via wireless external controller. the systems. They can also work equipment that uses the same “We studied the wireless commu- with signal-processing chips, noted frequency and communications pro- nications to understand the specifics Medtronic’s Dougherty. Some device tocol as the device. of how the IMD and [controller] com-

12 COMPUTER

municate and utilized that knowledge They must be careful that these ORNL researchers are also creating to send commands of our own to the approaches don’t block IMD func- new insulin-pump-system architec- IMD,” explained University of Wash- tionality or cause other problems tures, such as those that implement ington assistant professor Tadayoshi because that could necessitate surgi- encryption and those that better sup- Kohno. cally removing and then replacing a port important security properties “We were able to cause an device. like authentication. implantable cardiac defibrillator to St. Jude Medical Center facilities use emit a shock designed to induce [a proprietary approaches to secure their Encryption fatal heart rhythm],” said Fu. “The IMDs, including the St. Jude Medical Few IMDs encrypt signals, but this radio allowed us to listen to sample Accent RF pacemaker and the St. Jude will soon change, said CTG’s Moyle. radio communications between Medical Anthem RF cardiac-resyn- Encryption could limit data inter- a [controller] and the device, then chronization-therapy pacemaker. ception and hide the commands used replay the communication to control St. Jude officials wouldn’t comment with the devices so that only permit- the device.” on the nature of their techniques. ted controllers could work with them. The researchers also obtained Different types of IMDs could use However, noted Paul, there are sample patient information placed the same kinds of security approaches. limits to this approach because on the device, including name, birth However, those techniques might encryption capabilities could add date, and diagnosis. function differently with the vari- complexity and require more system In addition, they shut off stored ous types of IMDs. For example, an resources to function properly. Some settings in the IMD, which would have implanted insulin pump interacts with IMDs might not have sufficient battery left the device unable to respond to emergencies. Researchers are trying to meet the security IMD advances bring problems challenges that IMDs face. Most IMDs support only short- range communications, over distances from 2 to 5 centimeters. external components of the system and computing power to implement However, the radio technology is continually throughout the day. In certain encryption algorithms. improving. ORNL has communicated pacemakers, there is less interaction with IMDs at a range of 30 meters. and all components are internal. Zero-power defense “But this doesn’t mean that they are Adding complexity such as secu- intended to communicate at that dis- Oak Ridge research rity features could be undesirable tance,” said the lab’s Paul. Paul’s group at ORNL has con- because they could consume IMDs’ Longer ranges will enable greater ducted experiments implementing limited battery life. In response, patient mobility during in-home data attacks on commercially deployed researchers are considering a collection. In addition, the computer insulin-pump systems. security mechanism called the zero- equipment used to gather informa- In the process, they have devel- power defense. tion can be moved farther from the oped a detailed model of potential The goal is to enhance the secu- patient, thereby protecting sterile threats. For example, said Paul, if a rity of an IMD without using energy zones in operating and patient rooms. smart phone is used in insulin pump from the device’s battery, according However, the longer range will also systems in the future to store blood to the University of Massachusetts make IMD systems accessible to more glucose levels, as has been proposed, Amherst’s Ransford. people, including potential hackers. hackers changing those values could An energy-harvesting computer Meanwhile, as is the case with most hurt patients. could serve as a gateway device. devices, IMDs have become more dif- The researchers are addressing People trying to communicate with ficult to secure as they have become these threats by creating new proto- an IMD power the gateway device more functional and complex. cols to protect IMDs’ control channels with their own radio transmissions. and the patient data they store and The gateway then runs a challenge- FIGHTING BACK transmit. response protocol that makes people Researchers and manufactur- For instance, noted Paul, systems prove they’re allowed to contact the ers are trying to design security could use very-short-range commu- IMD. approaches that ensure that real- nications. This would make attacks Unauthorized parties are thus world attacks either don’t happen or more difficult by disallowing long- deterred without using any of the don’t cause problems. range attacks. IMD’s battery power, noted Ransford.

AUGUST 2010 13

TECHNOLOGY NEWS

Patient-centered approaches “Our observations suggest that But as IMDs become smaller, more University of Washington profes- no single security approach may be resource-constrained, increasingly sor Batya Friedman said the school’s attractive to all patients but rather complex, and more functional, the Value-Sensitive Design Research Lab, that different types of security challenges in making them secure which she codirects, surveyed cardiac approaches may appeal to different while taking into account consid- patients with IMDs about suggested patients,” Denning said. erations such as usability, patient security solutions. values, and battery life will increase. Friedman said patients preferred Issues “We really need a concerted effort security solutions that warned of IMD security faces several key on the part of all relevant stakehold- potential problems, didn’t require challenges. First, adding security ers, including computer security them to do anything inconvenient, could hurt system performance and researchers, medical practitioners, and didn’t call attention to their increase cost, at least initially. device manufacturers, [regulators], condition. Some approaches may require social scientists, and patient advocacy Some experts have suggested completely new devices or com- groups,” said the University of Wash- implementing passwords that must ponents. For example, healthcare ington’s Friedman. be entered before someone can providers implementing two-way This effort, added Paul, should also access an IMD. communications in IMD systems include the benefits of standardizing However, doctors who might not would have to replace unidirectional various IMD security properties. know the password would have to be equipment. Stated the University of Massa- able to control the devices in case of At a minimum, ORNL’s Paul said, chusetts’ Fu, “Legislators should give emergency, particularly if the patient adding technologies such as encryp- regulators the authority to require is unconscious. To deal with this, tion would require updates of the adequate privacy controls before an patients could wear bracelets that software on some IMDs and the IMD can reach the market.” show their passwords. However, they controller. Legislation should avoid mandat- could lose the bracelets. However, he added, the biggest ing specific technical approaches but One proposed solution popular challenge will be finding solutions instead just provide incentives and with the security community—IMD- that are acceptable to patients. penalties, he added. access passwords tattooed on patients However, he said, manufacturers as barcodes visible only under ultra- oday’s IMDs increasingly use are ultimately responsible for IMD violet light—met with mixed results wireless communications safety. because some respondents didn’t like that provide monitoring and “Medical devices save lives, but the idea of tattoos, explained Tamara other benefits for patients, they are no more immune to secu- Denning, a doctoral student at the so the technology is here to rity and privacy risks than any other Value-Sensitive Design Research Lab. stay,T said Paul. computing device,” he said. “We’d better get the security and privacy right during the early design stages because surgically replacing an inse- cure medical device is much less convenient than an automated Win- dows update. And the consequences can be fatal.”fi

Neal Leavitt is president of Leavitt Communications (www.leavcom. ___com), a Fallbrook, California-based international marketing communications company with affiliate offices in Brazil, France, Germany, Hong Kong, India, and the UK. He writes frequently on technology topics and can be reached at [email protected].

______Editor: Lee Garber, Computer; [email protected]

14 COMPUTER

NEWS BRIEFS Project Converts Arms and Hands into Input Devices

Carnegie Mellon Univer- found on mobile devices. Their size as Figure 1 shows. The system detects sity doctoral student has comes at the expense of usability, he the small vibrations the touches developed a prototype explained. make; recognizes the gesture; and A system that could let users Skinput and additional systems issues the associated command, such turn their arms or hands Harrison is working on use arms, as answering a phone or selecting an into keyboards or display screens via hands, and other larger areas near interface menu item. acoustic vibrations produced by tap- mobile devices for input and output, Users wear an armband—the ping their skin. thereby increasing usability and prototype is made with an elbow Chris Harrison, who is working interactivity. brace—lined with 10 specialized with Microsoft on the project, said his With Skinput, a person interacts sensors that detect the vibrations Skinput system would be preferable with devices by using a finger to touch and determine the action to be to the small keyboards and screens other fingers, their hand, or their arm, carried out. The sensors are canti-

Figure 1. A Carnegie Mellon University project turns users’ arms and hands into input devices for activities such as answering a smartphone or selecting an interface menu item. Users could, for example, utilize an arm or hand as a keyboard or display screen via acoustic vibrations produced by tapping the skin. Source: Chris Harrison, Carnegie Mellon University

NEWS BRIEFS

levered piezo films. These are thin vibration. In addition to enabling tasks on mobile devices, Skinput films—fixed to the brace at one end the system to recognize specific uses a tiny projector to display a and unattached at the other—acti- gestures, this lets it disregard extra- virtual keyboard with large keys on vated by electricity generated when neous noise. the user’s hand or arm. The system they’re subject to the mechanical The sensors connect to a PC—run- recognizes the keys pressed based stress that the vibrations cause. ning custom software—that performs on the resulting audio vibrations. So Weights added to the film’s loose the system’s necessary computations. far, researchers have only tested the end change the sensors’ resonant fre- In the future, the researchers hope system with a five-key board. quency so that they can detect small their system could also connect to a Microsoft researcher Dan Morris audio vibrations in the frequency smartphone or other mobile device to said Harrison began developing Skin- range—from 25 to 78 Hz—that propa- conduct the computations. put while an intern at the company. gate naturally up human bones and Users need only a few minutes He noted that Microsoft is working on soft tissues. to train Skinput to learn the audio other, similar systems. According to Harrison, the sen- vibrations peculiar to their gestures Skinput—still in early devel- sors used are like guitar strings: and to customize the commands that opment—won’t be commercially each is sensitive to a particular each gesture represents, explained available for up to seven years, audio frequency. Thus, when a user Harrison. Morris noted. It will need various taps a finger on a wrist, only one To help users send text messages improvements, including more accu- sensor will react to the resulting or otherwise navigate complex input rate sensors, he explained. Device Could Eliminate Wires in Home and Office Communications

urdue University research- the system converts the optical signal other approaches, the Purdue system ers are working on a small to a radio signal. yields transmissions in the 60-GHz device that converts laser The system works with a mode- frequency band without signal-com- P pulses into radio signals, locked laser that emits an optical bining techniques that create jitter. The which could enable high- pulse that lasts only 50 femtoseconds 57-to-64-GHz frequency band is desir- speed wireless communications in (50 quadrillionths of a second) but able because governments permit it for place of many of the wired transmis- that offers high bandwidth. unlicensed utilization globally, mean- sions currently used in home and A spectral shaping chip selects ing the systems would be usable and office systems. a few parts of the optical pulse and interoperable worldwide. With this chip-based, photonically adjusts the strength of each. The According to Qi, his team is work- assisted, radio-frequency, arbitrary microring resonators, the chip’s ing on incorporating the device’s waveform-generation device—devel- major components, pass information components—the laser, spectral- oped by a team led by Purdue about the shaped signals to a spool of shaping chip, and fiber spool—onto assistant professor Minghao Qi—a optical fiber. a single chip to make the system single base station could handle The fiber allows the different smaller and less expensive. wireless communications for dif- wavelengths of light within a trans- The US National Science Foun- ferent types of applications such as mission to travel at different speeds, dation, Defense Threat Reduction high-definition TV, detached laptop thereby forming the peaks and Agency, and National Security displays, or wireless printing servers. valleys of the signal formed after Science and Engineering Faculty Qi’s device uses shaping techniques photodiodes convert the light into a Fellowship program are funding the on the light signals—a technique that radio-frequency format. Purdue research. doesn’t work with radio signals—to The Purdue system is smaller Purdue has filed a provisional create more complex waveforms that and thus more portable than other patent for the technology, which carry more data and offer more band- approaches for converting laser pulses won’t be ready for commercializa- width, capabilities that remain when into radio signals. In addition, unlike tion for at least five years, Qi said.

16 COMPUTER

APPLICATION USES VISUALIZATION TECHNOLOGY TO FIGHT DISEASE

researcher has developed visualization techniques and a tool- reports contain dečned čelds such as time of day or age of perpe - A kit that let users comb through disease-related statistics in trator. But they also contain narrative with information not as otherwise unexamined sources—such as patient records and news- clearly dečned as and not structured like a table or graph. paper articles—to discover geographic trends that could help To create structure and derive meaningful conclusions from these control the spread of illnesses. unstructured documents, GeoViz lets researchers use text analytics, By using visual information representations such as interactive which lets them extract relevant information, Hardisty said. maps and graphs in combination with text searches, users could Mapping technologies let researchers organize, analyze, and more easily see relationships between data that provide insights present the information geographically. And GeoViz enables them into complex events like the spread of an illness, explained Frank to change the factors they look at and the perspective from which Hardisty, a research associate at Pennsylvania State University’s they analyze them GeoVISTA Center. GeoVISTA is an interdisciplinary geographic- As Figure A shows, Hardisty’s toolkit provides users with an information-science center. array of display options, including maps, parallel coordinate plots, Hardisty said geographic visualization could help analysts čnd and multiform matrices. This could show patterns in how and how previously unknown relationships among data. quickly a disease is spreading. The GeoViz Tookit, for which Hardisty was lead developer, could The analysis could also show locations that are susceptible or work in any čeld—such as public health or law enforcement—with not susceptible to certain diseases, as well as possible links to large amounts of publicly available data. The toolkit uses well-known causes or triggers. In addition, it could help oćcials identify at-risk data-mining and other algorithms, as well as open source software populations and implement preventative plans such as regular dis- libraries such as the Apache Lucene search-engine library. ease exams. GeoViz can be used by even those with no programming experi- Moreover, GeoViz could help public health oćcials develop ence, which makes it more useful than many other analytical tools eective control measures—like immunization regimens—for for researchers, Hardisty noted. both infectious and chronic diseases. Numerous potential sources for disease, crime, and other statis- Hardisty says his group is working with analysts at the US Cen- tics, such as newspaper articles, are in a semistructured format, ters for Disease Control and other health organizations that could with data that contains both numbers and text. For example, police implement his tool.

Figure A. A Pennsylvania State University researcher has developed the GeoViz Toolkit. GeoViz would enable users to analyze health-related statistics in otherwise unexamined sources—such as patient records and newspaper articles—to discover and display geographic trends that could help control diseases’ spread. Source: Frank Hardisty, Pennsylvania State University.

AUGUST 2010 17

NEWS BRIEFS Online Threat Extorted Porn Downloaders

olice have arrested a Jap- hackers’ desire to continue using phone number, mailing address, anese gang that allegedly social-engineering-based attacks employer name, and passwords used ran an extortion scheme rather than solely relying on technical to access games. P with a new pornographic exploits, noted Johannes B. Ullrich, It also automatically collected data twist. chief technology officer for the SANS from victims’ computers, including The hackers reportedly first Institute, a security training and information about personal accounts, fooled users trying to download information organization. browser histories, and Internet adult games via peer-to-peer (P2P) In the recent attacks, hackers Explorer favorites. technology into installing malware. reportedly seeded a Trojan called The Trojan uploaded all the infor- They then allegedly collected money Kenzero—disguised as an installer— mation to the hackers. They allegedly from the victims in exchange for into some adult games available via posted the information—along with removing online posts that included the popular Japanese Winny P2P screenshots of Internet Explorer their names, personal data har- network, explained Rik Ferguson, a Favorites and records of computer vested from their computers, and senior security researcher with secu- activity such as access of the por- information about their attempted rity vendor Trend Micro. nographic games—with the victim’s pornographic downloads. Downloading the anime-style name to a website that the host ISP This attack is significant because games, known as Hentai, isn’t illegal. has since taken offline. it demonstrates that cybercriminals But the hackers may have exploited A party identifying itself as have become aware of and are taking the potential shame some people Romancing Inc. then sent e-mails advantage of users’ P2P activities, would have if others found they did to victims claiming that they down- said Rossano Ferraris, functional so, Ferguson noted. loaded the game in violation of lead for Internet security intelligence When victims tried to download someone else’s copyright and that with CA Technologies’ Internet Secu- a game affected by the scheme, the their name and the potentially embar- rity Business Unit. installer/Trojan requested personal rassing information would remain on The attack also demonstrates data, including name, birth date, the Web until they resolved the issue. The hackers allegedly offered to remove the information and resolve the alleged copyright infringement for a fee reported in some cases to be about 1,500 yen (currently about $17), paid by credit card. COMPUTING According to reports in the Japa- nese media, at least 5,500 people downloaded the malicious file. Law enforcement didn’t release details THEN about how many people actually paid the criminals. The only victims identified were in Japan. Learn about computing history Although cybercriminals have used copyright infringement scams and the people who shaped it. before, the pornographic angle is new, according to Ullrich.

http://computingnow.______News Briefs written by Linda Dailey Paulson, a freelance technology computer.org/______ct writer based in Portland, Oregon. Contact her at [email protected].

Editor: Lee Garber, Computer; [email protected]______

18 COMPUTER

PURPOSE: The IEEE Computer Society is the world’s largest EXECUTIVE COMMITTEE association of computing professionals and is the leading President: James D. Isaak* provider of technical information in the field. President-Elect: Sorel Reisman* Past President: Susan K. (Kathy) Land, CSDP* MEMBERSHIP: Members receive the monthly magazine VP, Standards Activities: Roger U. Fujii (1st VP)* Computer, discounts, and opportunities to serve (all activities Secretary: Jeffrey M. Voas (2nd VP)* are led by volunteer members). Membership is open to all IEEE VP, Educational Activities: Elizabeth L. Burd* members, affiliate society members, and others interested in the VP, Member & Geographic Activities: Sattupathu V. Sankaran† VP, Publications: David Alan Grier* computer field. VP, Professional Activities: James W. Moore* COMPUTER SOCIETY WEB SITE: www.computer.org VP, Technical & Conference Activities: John W. Walz* OMBUDSMAN: To check membership status or report a change Treasurer: Frank E. Ferrante* 2010–2011 IEEE Division V Director: Michael R. Williams† of address, call the IEEE Member Services toll-free number, 2009–2010 IEEE Division VIII Director: Stephen L. Diamond† +1 800 678 4333 (US) or +1 732 981 0060 (international). Direct 2010 IEEE Division VIII Director-Elect: Susan K. (Kathy) Land, CSDP* Computer all other Computer Society-related questions—magazine delivery Editor in Chief: Carl K. Chang† or unresolved complaints—to [email protected].______* voting member of the Board of Governors † nonvoting member of the Board of Governors CHAPTERS: Regular and student chapters worldwide provide the BOARD OF GOVERNORS opportunity to interact with colleagues, hear technical experts, Term Expiring 2010: Piere Bourque; André Ivanov; Phillip A. Laplante; and serve the local professional community. Itaru Mimura; Jon G. Rokne; Christina M. Schober; Ann E.K. Sobel AVAILABLE INFORMATION: To obtain more information on any Term Expiring 2011: Elisa Bertino, George V. Cybenko, Ann DeMarle, David S. Ebert, David A. Grier, Hironori Kasahara, Steven L. Tanimoto of the following, contact Customer Service at +1 714 821 8380 or Term Expiring 2012: Elizabeth L. Burd, Thomas M. Conte, Frank E. +1 800 272 6657: Ferrante, Jean-Luc Gaudiot, Luis Kun, James W. Moore, John W. Walz H Membership applications EXECUTIVE STAFF H Publications catalog Executive Director: Angela R. Burgess H Draft standards and order forms Associate Executive Director; Director, Governance: Anne Marie Kelly Director, Finance & Accounting: John Miller H Technical committee list Director, Membership Development: Violet S. Doan H Technical committee application Director, Products & Services: Evan Butterfield H Chapter start-up procedures Director, Sales & Marketing: Dick Price H Student scholarship information COMPUTER SOCIETY OFFICES H Volunteer leaders/staff directory Washington, D.C.: 2001 L St., Ste. 700, Washington, D.C. 20036 H IEEE senior member grade application (requires 10 years Phone: H-D practice and significant performance in five of those 10) Email: [email protected]______Los Alamitos: 10662 Los Vaqueros Circle, Los Alamitos, CA 90720-1314 PUBLICATIONS AND ACTIVITIES Phone: +1 714 821 8380 Email: [email protected]______Computer: The flagship publication of the IEEE Computer Society, Membership & Publication Orders: Computer, publishes peer-reviewed technical content that Phone: H-D covers all aspects of computer science, computer engineering, Email: [email protected]______technology, and applications. Asia/Pacific: Watanabe Building, 1-4-2 Minami-Aoyama, Minato-ku, Tokyo 107-0062, Japan Periodicals: The society publishes 13 magazines, 18 transactions, Phone: H-D and one letters. Refer to membership application or request Email: [email protected]______information as noted above. IEEE OFFICERS Conference Proceedings & Books: Conference Publishing President: Pedro A. Ray Services publishes more than 175 titles every year. CS Press President-Elect: Moshe Kam publishes books in partnership with John Wiley & Sons. Past President: John R. Vig Standards Working Groups: More than 150 groups produce Secretary: David G. Green Treasurer: Peter W. Staecker IEEE standards used throughout the world. President, Standards Association Board of Governors: Technical Committees: TCs provide professional interaction in W. Charlston Adams more than 45 technical areas and directly influence computer VP, Educational Activities: Tariq S. Durrani VP, Membership & Geographic Activities: Barry L. Shoop engineering conferences and publications. VP, Publication Services & Products: Jon G. Rokne Conferences/Education: The society holds about 200 VP, Technical Activities: Roger D. Pollard conferences each year and sponsors many educational activities, IEEE Division V Director: Michael R. Williams IEEE Division VIII Director: Stephen L. Diamond including computing science accreditation. President, IEEE-USA: Evelyn H. Hirt Certifications: The society offers two software developer credentials. Next Board Meeting: For more information, visit www.computer.org/certification. 15–16 Nov. 2010, New Brunswick, NJ, USA

revised 17 Jun. 2010

PERSPECTIVES

Microprocessor Scaling: What Limits Will Hold?

Jacek Izydorczyk and Michael Izydorczyk, Silesian University of Technology

For 40 years, semiconductor technology has followed Moore’s law, but continuing that pattern will require a breakthrough in energy-efficient design. Eventually the industry must also seek an entirely new paradigm, such as creating reliable systems from unreliable components.

ature imposes fundamental limits on systems from unreliable components. Toward the end of computer design in the form of ther- the next 40 years, we also expect energy-efficient analog modynamic laws and restrictions from processing, which would enable computing machines with quantum mechanics. From the first an architecture that is remarkably similar to the human integrated circuit (IC) design to modern nervous system. Nsystems on a chip, designers have had to account for these Our research and the results of other experiments in some form. In addition, technological evolution has have led us to conclude that the capabilities of future been governed by Moore’s law—IC cost decreases, and technology will depend not on the size of integrated the number of transistors on an IC doubles approximately devices but on the IC’s ability to drain dissipated heat. every two years. Thus, we see two main problems: how to reduce the Lately, however, people have begun to question Moore’s energy needed to process one bit of data and how to law and the validity of long-accepted technology develop- drain heat from the chip. To seek answers, we’ve exam- ment metrics in future efforts. Will the pattern slow or ined Moore’s law, the hierarchy of limits on technology escalate? We believe that technology, in terms of bits pro- development, thermodynamic laws, and information cessed per second per square centimeter of IC real estate, theory. Our exploration has led us to make some inter- will develop exponentially in the next four decades. The esting predictions. first 20 years or so will be dedicated to finding new materials, devices, and circuits to enhance energy-efficient data PROGRESS ACCORDING TO MOORE processing, but the decades that follow must focus on new Many of our deductions are based on the past evolution design paradigms, such as creating reliable and predictable of complementary metal-oxide semiconductor (CMOS)

THE BIRTH OF MOORE’S LAW

technology according to established laws, primarily n 1954, Gordon Earle Moore was a doctoral candidate in physical Moore’s law. “The Birth of Moore’s Law” sidebar describes I chemistry at the California Institute of Technology and wrote his dissertation on nitrogen’s infrared spectroscopy in physical the story behind this enduring principle. chemistry. Shortly after, William Shockley hired Moore to work at Initially, in the 1960s Gordon E. Moore looked at semi- Shockley Semiconductor,1 which was his črst hands-on experience conductor improvement in four ways:1 with semiconductor devices. Moore and his colleagues, including Robert Noyce, designed the technology to produce silicon t size of the silicon wafers, transistors. Unfortunately, Shockley’s research was aimed at improving an exotic four-layer diode, a vision that clashed with that t overall miniaturization of the technology, of his younger coworkers. t imagination of those allocating devices on the wafer, Consequently, in 1957, Moore, Noyce, and six others (the infa- and mous Traitorous Eight) broke away and started up Fairchild t innovation in circuit construction. Semiconductor, with Moore quickly becoming the company’s main technologist and a key manager. The goal was to produce transistors from silicon instead of the then popular germanium, with the However, after Moore and Robert Noyce founded Intel idea that the devices would be disposable after a certain period. and began churning out microprocessors and semicon- In 1960, the company built a circuit with four transistors on a ductor memories, Moore changed the way he viewed single silicon wafer, creating the črst silicon IC. By the mid-1960s, ICs improvement. At first, progress was largely a matter of were competing with circuits assembled from discrete elements, so rapidly making smaller and smaller components, but in IC manufacturers began concentrating on minimizing the fabrication cost of a single device inside the IC. Because almost all IC an extremely short time, the entire semiconductor indus- production cost stemmed from circuit development, the cost of fab- try went digital and the market opened for advanced ICs, ricating a single device decreased as the number of devices on the which changed the design focus. Improvement began to IC increased. be measured in the number of transistors that could be However, with more devices per IC came the problem of surface integrated on an IC, which in turn depended on reducing area on the silicon wafer: the larger the surface, the greater the exposure to defects and the lower the yield. Consequently, the goal fabrication dimensions. This new design focus slowed became to čnd an optimal IC size that guaranteed a minimal single the improvement rate, and Moore ventured a predic- device cost at an acceptable performance level. As companies tion: the number of integrated transistors would double began fabricating more ICs in line with this goal, Moore noted a pat- about every 26 months.2 Moore’s prediction might have tern, which he črst documented in 1965. 2 The minimal device cost remained within Intel’s walls were it not for publications decreased for circuits that contained more transistors, and the number of transistors on one circuit had doubled every year since like Science and Scientific American, which popularized the construction of the črst IC in 1960. Moore later revised the dou - 1 his and Royce’s articles. bling period to every two years, a pattern that remains valid today In retrospect, it’s hard to say with certainty if Moore and is well known as Moore’s law. was prescient or if he was merely outlining a development roadmap with an eye toward the already fierce References 1. C. Lécuyer and D.C. Brock, “Biographies: Gordon Earle Moore,” market competition and capacity for technical innovation. IEEE Annals of the History of Computing, vol. 28, no. 3, 2006, pp. Whether or not it was a self-fulfilling prophecy, Moore’s 89-95. law has held since it first became known as such. Only 2. G. Moore, “Cramming More Components onto Integrated Cir- recently have people begun to ask how much longer this cuits,” Electronics, vol. 38, no. 8, 1965, pp. 114-117. law will remain valid.

HIERARCHY OF LIMITS law of information theory, for example, governs informa- To answer that question, we turned to the hierarchy tion transmission, limiting the efficiency of information of constraints on computing machines proposed by J.D. processing, such as the amount of energy to process one Meindl et al. in the late 1980s.3 At the top are fundamen- bit of data. tal limits that incorporate the most general principles Other examples of these limits are that signals cannot of efficiency; at the bottom are circuit limits specific to propagate at a speed greater than the speed of light in CMOS technology. a vacuum, and the machine’s smallest parts can’t vio- late quantum mechanics. Thus, particles must obey the Fundamental limits uncertainty principle; electrons are limited by the Pauli Any reasonable computing machine must be in thermo- exclusion principle; and so on. dynamic balance with its environment, following certain laws that govern its efficient use of heat or other energy Material limits forms. Entropy, which is the disorder that thermodynamics These limits arise from the properties of the semicon- or statistical mechanics research must consider, is inher- ductors, metals, and insulators used in IC production, ently the same for all kinds of computing systems. The which constrain a computing machine’s dimensions and

AUGUST 2010 21

PERSPECTIVES

V Logic gates Threshold dd T1 Computations are thermodynamically irreversible InputEvaluation AWGN Output Input Output because computers inevitably dissipate energy; a micro- tree channel processor must transfer heat to the ambient environment. (a) (b) T2 Overall heat transfer is strictly related to temperature, a core concept in thermodynamics. Entropy, which is basically the amount of data processed, is also a key idea Figure 1. Limits of logic gates. In this model of (a) a three- in thermodynamics. In logic gate description in ther- stage logic gate and (b) a CMOS inverter, the machine could do modynamics, as well as in information theory, entropy computations without energy dissipation, but it must spend inuences the limits that designers must consider. energy to read out the result under the constant impact of thermal fluctuations. To illustrate the limits imposed by logic gates, we offer the model in Figure 1a,7 which has three cascaded stages. The first stage is the evaluation tree, which is a mecha- processing power. Researchers agree that material limits nism to compute the logic function’s value without energy arise from four obstacles: switching energy, transit time, dissipation. thermal conductance, and dopant uctuation. The next stage is the additive white Gaussian noise (AWGN) channel, which describes the process of read- Device limits ing the output from the evaluation tree. The AWGN adds Device limits can be traced to the continued use of bulk Gaussian, mutually uncorrelated noise values to the data metal-oxide-semiconductor field-effect transistors (MOS- transferred from input to output. Information theory treats FETs) as switching devices. Problems with current leakage the AWGN channel as a basic model of a real system that and capacitive coupling between a gate and a channel transfers data in the presence of noise. are motivating researchers to study alternative solutions, The third stage is a threshold element that restores such as a fully depleted silicon-on-insulator MOSFET with binary data read out from the evaluation tree. We include multiple gates. this stage only to complete our examination of thermodynamic principles. We don’t view it as necessary to a Circuit limits logic-gate model because performing this function imme- CMOS logic circuits primarily dictate circuit limits, diately after receiving input isn’t an optimal strategy. and these circuits still suffer from latency and signal Figure 1b shows a basic CMOS inverter. Even this contamination when interconnected. Other problems basic form reveals elements of the model in Figure 1a. include performance uctuation and yield, both of which The induction of a channel under transistor T1’s gate and are of extreme interest in light of emerging logic tech- the depletion of that channel under transistor T2’s gate nologies. So far, most researchers don’t believe that these determine a function’s value. Alternatively, the value is alternative technologies have much promise to improve computed by the voltage applied to the transistors’ gates. performance.4 It’s possible to read the logic machinery’s state only One potential limit buster is the introduction of material when the current ows through one of the transistors, with a high dielectric constant, or high-g, for gate insula- either T1 or T2, and loads or unloads a parasitic output tion. Although some researchers didn’t believe the material capacitor. Charging the capacitances leads to an energy would be introduced in the next CMOS generation,5 toward accumulation, which discharging dissipates. the end of 2007 Intel started commercial production of the Although common sense says that energy dissipa- 45-nm Penryn processor based on high-g technology.6 The tion is mandatory, in reality it isn’t essential for CMOS Intel Core i3, i5, and i7 microprocessors have successfully logic, a principle illustrated in adiabatic logic gates. implemented this processor. These gates are an example of how changes in the circuit alone, without new devices or novel materials, can CHALLENGES AT THE BOTTOM greatly reduce energy dissipation. Adiabatic logic has Generally, the higher the limit in the hierarchy, the more certain drawbacks, and it isn’t certain how prevalent this chance that technological progress will bust it. Thus, we kind of circuit will become. However, it is proof that the believe that lucrative breakthroughs won’t be at the higher charging-discharging cycle for energy dissipation isn’t levels. Rather, we concur with Richard Feynman, who a requirement. remarked “there’s plenty of room at the bottom,” mean- It also implies that only thermal noise disturbs this ing that big profits are at the lowest level, where creative cycle and that a certain amount of this noise is unavoid- design working within the laws of thermodynamics and able. The AWGN channel in Figure 1a accounts for the idea quantum mechanics can reap rewards—and where energy that an infinitesimally low voltage ramp cannot charge efficiency should be the primary goal. capacitances. The AWGN channel not only introduces

22 COMPUTER

noise but also reects the band limit dictated by the cur- Source Gate Drain rent ow’s dynamics. Maxwell Thermal noise and energy dissipation statistics Fermi To understand why thermal noise is unavoidable, con- U level sider the simple one-dimensional model of an n-channel Band gap field-effect transistor (FET) in Figure 2. A FET is the basic Fermi device for current CMOS technology and likely to remain statistics a building block in future technologies such as carbon nanotubes. The two energy wells in the figure denote the source and drain, which hold the electrons responsible for Figure 2. Simple model of an n-channel field-effect transistor conductivity. The gate is a high-energy region between the (FET). The source and drain hold the electrons responsible for source and drain. Electrons are subject to Fermi-Dirac sta- conductivity. Below the Fermi level, electrons fill almost all tistics—a law that governs how a crowd of electrons must possible energy states. Above the Fermi level, many states are behave. Below the Fermi level, electrons are crowded, fill- empty and the electrons are free, described by Maxwell statistics. U ing almost all possible energy states; above the Fermi level, If a voltage difference ( ) exists between the source and the drain, high-energy electrons will enter the drain region, where they will many states are empty and the electrons are free. remain (localized). Dopant concentration in the source and drain region is usually high enough (1015 to 1020 cm<3) that the Fermi level is either just below the bottom of the conduction band or bit of data at room temperature (T = 300 K) is extremely <7 even a little above it. If the potential energy of electrons in small (¡b 17.9 meV, where 1 eV = 1.602 = 10 pJ). the gate region is sufficiently large, the source and drain are separated. Quantum mechanics restrictions Decreasing the energy barrier in the gate region allows Transistor dimensions are restricted by the product some electrons to surmount the barrier and thus be any- of particle momentum dispersion and particle position where in the transistor. These escaped electrons have an dispersion, which in turn is bounded by the uncertainty energy level high enough to generate quantum leakage principle. Thus, any attempt to localize the particle in through the barrier. As Figure 2 shows, if a voltage differ- the tightly bounded position can enlarge the particle ence (U) exists between the source and drain, high-energy momentum disturbance. When a single electron is con- electrons will enter the drain region. Once they interact fined in the potential well, the uncertainty principle with other electrons and the crystal lattice, which lowers manifests itself as the constant product of an energy their energy level, they’ll remain in that region. Even if the gap between the permitted energy levels and squared electrons scatter free over the barrier, some mechanism is well width. Energy levels inside a wide well are very still needed to dissipate their energy so that they remain close, and a scattered particle causes them to merge into in the drain region. Otherwise, electrons reect from the a single band. rear drain wall, much like a Ping-Pong ball bounces off a In contrast, energy levels in a very narrow well are hard oor. At first the ball is attracted to the hard oor by wide apart and thus far less likely to overlap, even when gravity, but then it consistently bounces high after collid- the particle interacts with the crystal lattice. A small FET ing with it. naturally dissipates energy while the electron is local- Trapping electrons in the drain region in this way is an ized. An extremely small device—about 5 nm wide—can effective strategy when an energy increase caused by the dissipate the excess energy, but a larger device—about 28 move from source to drain significantly changes the aver- nm wide—translates to an energy gap on the order of the age number of electrons in the state more than once. To Landauer limit. illustrate, assuming that Maxwell statistics and the transfer To illustrate these restrictions, consider a hypothetical of one electron charge (e 1.602 = 10<19 C) from the source IC with the following characteristics: to the drain are sufficient to change the gate’s state. The energy dissipated is then t energy dissipated per processed bit is as low as the

Landauer limit at room temperature (¡b 17.9 meV);

eU = kTln2 = ¡b, t it uses planar technology; t its modus operandi is one electron device; where k is the Bolzmann constant of approximately t quantization of electron energy allows minimal linear 1.38 = 10<23 J/K and T is the FET temperature in Kelvin. dimensions; and Thus, the equation above represents the ultimate Landauer t it has a switching frequency of 3.5 GHz, consistent limit. The lower limit of the energy spent to process one with the top commercial microprocessor line.

AUGUST 2010 23

PERSPECTIVES

of 1,012 switching elements per square centimeter and Ultimate Shannon limit operate at 8 MHz only. 10 AWGN with binary input Binary transmission STAGES OF ENERGY EFFICIENCY We predict that the work to create energy-efficient 1 technologies will evolve in three stages: a new metric for integration density and processing speed, a new design paradigm that allows the construction of reliable machines 0.1 from unreliable components, and energy-efficient analog Stage 1 data processing. Figure 3 shows the three stages. Stage 3 (~2 dB) Stage 2 (~17 dB) (~26.6 dB) Data rate (mean number of bits per waveform) 0.1 1 10 100 A new density metric Log of energy to process one bit lower limit of energy to process one bit The current estimated lower bound on energy used to process one bit is approximately 647 eV, about 36,000 Figure 3. Three evolutionary stages of energy-efficient times higher than the absolute Landauer limit. Unfortu- technology. The horizontal axis represents the energy spent to nately, to achieve the Landauer limit the design must use process one bit of data on a double logarithmic scale moving from advanced coding and decoding algorithms. the Landauer limit to infinity. The vertical axis represents data rate At present, digital design typically consists of a binary as the mean number of bits carried by one waveform. The ultimate symmetrical channel organized along the lines of a Gauss- Shannon limit divides the plane into two domains. In the upper ian channel and with bipolar binary modulation. The domain, reliable transmission and data processing are impossible. optimal demodulator would be a matched filter, sampler, The lower domain contains the additive white Gaussian noise and threshold element that maximizes the power ratio (AWGN) for binary transmission and with binary input. of useful signal to thermal noise and thus minimizes the probability of transmission error. More than 400 W/cm2 of heat density must be removed Optimizing the decoder will become important. Current from the surface of this hypothetical chip—which is hard logic circuits aren’t built around the concept of an optimal to achieve outside a research lab. Assume that the IC tem- decoder. However, the minimal energy needed to process perature is higher than the ambient temperature and less one bit of data without coding bits, a value of 1.4 eV, is only than 400 K. Theoretically, the best planar-circuit cooling weakly related to the heat stream density removed from method would be forced liquid ow and boiling, which the chip and the acceptable error ratio.7 And heat stream can drain up to 1 kW/cm2 of heat from the chip.8 For 3D density is key to energy efficiency as it cannot exceed structures, the theoretical amount that can be drained is 300 W/cm2. even higher, about 10 kW/cm2. The heat stream drained from the IC—the energy spent However, in practice neither this method nor a heat to process one bit, multiplied by the number of bits pro- stream formed from thermoelectric phenomena will cessed per second—isn’t likely to change. For the past two make a significant difference in the chip’s 300 W/cm2 to three decades, a typical general-purpose microproces- border.8 Radical progress in cooling technology seems sor has consistently used slightly more than 1 cm2 of its unlikely. As the sidebar “The Impracticality of Cryogen- silicon wafer. The energy spent to process one bit is now ics” describes, cryogenics has long been a source of about 647 eV, and number of bits processed per second per hope, but practical concerns have kept it from commer- 1 cm2 of IC is now about 1018. cial development. Energy dissipation must come from All this argues for a new processing-speed metric. The another source. latest trends show a systematic decrease in area with an One possibility is to reduce the energy spent processing asymptotic value of 1 cm2. Assuming that new value, we data. To put the problem in perspective, the Intel Core2 believe it’s possible to estimate a microprocessor’s total (four cores) has a total power dissipation of 136 W with processing power by calculating the number of bits pro- 820 million switching transistors and a 3.2-GHz switching cessed in one second on 1 cm2 of wafer space. This value frequency (www.intel.com/products/desktop/processors/ could then replace the number of integrated devices as

index.htm?iid=processors_body+dt_core).______Assuming a the metric of integration density and processing speed. conservative average of two transistors to process one If in the near future heat stream density increases to 300 bit of data, the value of energy per bit is approximately W/cm2, and assuming that the practical limit of energy 647 eV. Obviously this value must greatly decrease, but dissipated per bit remains 1.4 eV, an integration density of integrating as many transistors as possible on a chip approximately 1.3 = 1021 bits per cm2 per second should works against that. With switching frequency unlikely to be possible—a 1,300-fold increase in processing power. improve, the most densely integrated circuit will consist In Figure 3, this first evolutionary stage spans about

24 COMPUTER

26.6 dB on the energy scale, and the design preserves the THE IMPRACTICALITY OF CRYOGENICS equivalence between waveform and bit, or wave-to-bit correspondence, with a reasonable error rate. lthough cryogenic devices were initially promising, oćering the A potential for ultrahigh switching speeds, operating large A new design paradigm equipment at temperatures close to absolute zero was an If the integration density growth rate that results from unsolvable problem. If the IC is cooled to cryogenic temperatures, uctuations caused by thermal noises drop drastically, and hence Moore’s law remains constant, we anticipate about 22 less energy is needed to process one bit of data. Unfortunately, the years of uninterrupted development. At the end of this requirement to keep the circuit at a temperature below the ambient time, we expect to see the next evolutionary stage. Ad temperature requires some cooling mechanism. A cryogenic hoc methods similar to those described by Robert P. Col- computer dissipates little heat, but energy must be dissipated to well, chief architect of the P6 microprocessor family, can’t drain oć heat to keep the circuit below ambient temperature. And the eciency of any cooling machine is worse than the eciency of alleviate soft errors caused by thermal noise. In The Pen- a Carnot cycle. Consequently, the total dissipated energy for tium Chronicles, Colwell describes transient errors in the computation and cooling is no lower than the heat dissipated Pentium 4 machines: “. . . forward progress of the engine during computation at ambient room temperature. is self-monitored, and if too much time has elapsed since Moreover, “cold” electrons are associated with very long waves, forward progress was last detected, a watchdog timer and the device size limit is close to what devices are already. The switching time of such a circuit remains at 1.4 ps, but the heat den- will ush the machine and restart it from a known-good sity is only 43 W/cm2, which allows the construction of an IC with a 9 spot. . . .” switching frequency in the hundreds of gigahertz. This brief description underlines the need for a rev- These realities make cryogenic technologies very resistant in olution in circuit design. If indeed no device is reliable practical applications, which prompted Niklaus Wirth, Swiss com- because of transient errors, then the design paradigm puter scientist and creator of Pascal, to include them in a group of elegant but not too sensible ideas about computing.1 In his retro- should shift to reliable, predictable systems from unreli- spective of cryogenics, Wirth noted that, with the appearance of able components with unpredictable behavior. personal computers, cryogenic dreams were expected to “either Designers use two fault-tolerant techniques to increase freeze or evaporate,”1 and nothing on the horizon promises to a system’s reliability, availability, or dependability. The change that outlook. first is to select highly reliable components. The second is to use protective redundancy, just as communication Reference uses a repeat code. However, neither of these techniques 1. N. Wirth, “Good Ideas, through the Looking Glass” Computer, vol. 39, no. 1, 2006, pp. 28-39. is the path to the envisioned paradigm shift. A more promising approach is to implement techniques akin to advanced forward-correction codes at all levels of the system hierarchy, starting with circuits that implement limit restricts the final exploration of analog computations. base arithmetic operations such as add, shift, complement, Even so, the huge discrepancy between the data rate of and multiply. binary transmission and the ultimate Shannon limit in Invented codes could have several properties simi- the first and second stages can support energy-efficient lar to those of communication codes, as well as having analog data processing. many distinct and unique differentiating properties. With Experimental results show that an analog coprocessor these codes, the minimal energy spent processing one bit used to solve ordinary differential equations dissipates no would be only slightly above the Landauer limit. Eventu- more than 1 percent of the energy of a general-purpose ally, the processing speed could reach 6.6 = 1022 bits per digital microprocessor and no more than 20 percent of cm2 per second without breaking the heat dissipation limit the energy of a digital signal processor. At the same time, of 300 W/cm2. If the technological growth rate resulting computations have been up to 10 times faster.10 Perhaps from Moore’s law continues, we foresee about 12 years of this is a harbinger of the analog computing era that so development. many researchers have predicted. In Figure 3, this second evolutionary stage spans about 17 dB on the energy scale. Wave-to-bit correspondence gradually relaxes, which probably reflects the greater e’re optimistic about the prospects number of devices processing one bit. for microprocessor scaling and see the possibility of exponential Analog data processing growth and the retention of Moore’s As Figure 3 shows, below the energy scale of 28.1 meV, law. Although no solution to energy further binary data processing is impossible, but analog Wreduction per bit or to cooling is on the horizon, the computation remains feasible. The small distance between continuous small upgrades of current CMOS technol- the limiting energy of the second stage and the Landauer ogy are evidence that no one can be certain of the

AUGUST 2010 25

PERSPECTIVES

technology’s limitations. We remain confident that such of Moore’s Law,” IEEE Trans. Very Large Scale Integration upgrades will continue to provide insights into new (VLSI) Systems, vol. 18, no. 1, 2010, pp. 161-165. directions for materials, devices, circuits, downscaling, 8. V.V. Zhirnov et al., “Limits to Binary Logic Switch Scaling— and interchip communication. As researchers solve A Gedanken Model,” Proc. IEEE, vol. 91, no. 11, 2003, pp. 1934-1939. new problems, they’ll push the limits of microproces- 9. R.P. Colwell, The Pentium Chronicles: The People, Passion, sor scaling until at some point a design paradigm shift and Politics behind Intel’s Landmark Chips, Wiley/IEEE CS will occur. Press, 2005. 10. G.E.R. Cowan et al., “A VLSI Analog Computer/Digital Com- References puter Accelerator,” IEEE J. Solid-State Circuits, vol. 41, no.1, 1. E. Mollick, “Establishing Moore’s Law,” IEEE Annals of the 2006, pp. 42-53. History of Computing, vol. 28, no. 3, 2006, pp. 62-75. Jacek Izydorczyk is an assistant professor of electron- 2. G. Moore, “Progress in Digital Integrated Electronics,” ics at Silesian University of Technology, Gliwice, Poland. Technical Digest—IEEE Int’l Electron Devices Meeting, vol. His research interests include network invariants, circuit 21, 1975, pp. 11-13. simulations, magnetic media modeling, signal process- 3. J.D. Meindl, Q. Chen, and J.A. Davis, “Limits on Silicon ing, speech synthesis, speech modeling, and analog filters. Nanoelectronics for Terascale Integration,” Science, vol. 293, no. 5537, 2001, pp. 2044-2049. Izydorczyk received a PhD in electronics from Silesian 4. J.A. Hutchby et al., “Emerging Research Memory and Logic University of Technology. He is a senior member of IEEE. Technology,” IEEE Circuits & Devices Magazine, vol. 21, no. Contact him at [email protected]. 3, 2005, pp. 47-51. Michael Izydorczyk is an undergraduate student study- 5. S. Luryi, J. Xu, and A. Zaslavsky, eds., Future Trends ing genetics at ETE in Gliwice, Poland. His research interests in Microelectronics, John Wiley & Sons/IEEE Press, include computer simulation, computational genomics, and 2007. bioinformatics. Contact him at [email protected]. 6. M.T. Bohr, R.S. Chau, and K. Mistry, “The High-g Solution,” IEEE Spectrum, vol. 44, no. 10, 2007, pp. 29-35. Selected CS articles and columns are available for free at 7. J. Izydorczyk, “Three Steps to the Thermal Noise Death http://ComputingNow.computer.org.

DEFEAT CYBER CRIMINALS. CYBERSECURITY AND YOUR COMPETITION.

Sharpen your skills and give yourself a major edge in the job market 9Zh^\cViZYVhVCVi^dcVa8ZciZgd[6XVYZb^X:mXZaaZcXZ^c with a cybersecurity degree from University of Maryland University >c[dgbVi^dc6hhjgVcXZ:YjXVi^dcWni]ZCH6VcYi]Z9=H College (UMUC). Our degrees focus on technical and policy aspects, 6YkVcXZYk^gijVahZXjg^inaVWZcVWaZhhijYZcihidXdbWVi preparing you for leadership and management roles—and making simulated cyber attacks you even more competitive for thousands of openings in the public and private sectors. Courses are available entirely online, so you HX]daVgh]^eh!adVchVcYVc^ciZgZhi"[gZZbdci]aneVnbZci can earn your degree while keeping your current job. plan available

Enroll now. 800-888-UMUC umuc.edu/______cyberwarrior

26 COMPUTER

COVER FEATURE

SOCIAL MULTIMEDIA COMPUTING

Yonghong Tian, Peking University Jaideep Srivastava, University of Minnesota Tiejun Huang, Peking University Noshir Contractor, Northwestern University

The explosive growth of social multimedia thereby creating asynchronous multimedia conversations. content on the Internet is revolutionizing Social multimedia also provides an additional context for content distribution and social interac- understanding multimedia content. For example, aggregat- ing behavioral data (such as click and pause) over all users tion. It has even led to a new research area, watching the same video might reveal the video’s most called social multimedia computing. interesting scenes or objects. Clearly, social multimedia has great potential to change how we communicate and collaborate.1 logs and social networks are becoming Computing technology has similarly evolved rapidly an increasingly important part of media over the past decade. Motivated by the growth of social consumption for Internet users. With the ubiq- media applications, social computing has emerged as a uitous presence of capture devices such as novel computing paradigm that involves studying and B phones, digital cameras, and camcorders, the managing social behavior and organizational dynamics to Internet has been transformed into a major channel for produce intelligent applications.2 However, the wide prev- multimedia content delivery. The next evolution is upon alence of social multimedia poses a significant challenge us, as the past decade has witnessed a coming together for social computing because many new issues involving of social networking sites—Facebook, MySpace, Blog- social activity and interaction around multimedia must be ger, LinkedIn—and content-sharing services—YouTube, addressed in a media-specific manner. Flickr, Youku—that have sprung up as platforms to facili- Nevertheless, multimedia research still remains open, tate users’ creating and sharing content as well as building given the challenging nature of this area’s research focus. large groups of friends. Social multimedia can help improve existing multimedia The hybrid of multimedia and social media, which we applications, so we use the term social multimedia comput- call social multimedia, supports new types of user interac- ing to denote the more focused multidisciplinary research tion. For example, YouTube recently introduced a feature and application field between social sciences and multi- that lets users respond to other users’ video contributions, media technology.

COVER FEATURE

Social computing Multimedia Human and technology social studies Two views of social multimedia computing Multimedia computing

An example Submitter

Tags

Discussion Image stats

Applying knowledge from social Using multimedia to study studies to design and improve social behavior and multimedia applications organizational dynamics

Sources Annotation: 1) Banquet? 2) Party? 3) Meeting? Tags Users a) Using collaborative tagging to infer the image semantics a) Social network based on collaborative tagging Mary Given: Jim is Mary is Jim Infer: Kate is who? Diana

b) Using social network to assist face recognition or image analysis b) Multimedia social network based on image analysis

Figure 1. Two views of social multimedia computing. The first shows social computing over multimedia, which centers on social sciences. The second shows social-empowered multimedia computing.

CHARACTERIZATION OF SOCIAL MULTIMEDIA group, and multimedia is the medium for how they Most likely, Mor Naaman first defined the social connect and transfer information. As such, text- multimedia in his blog (http://blog.radvision.com/ only blogs and SMS are social media but not social videooverenterprise/?p=172) as “an online source of multi- multimedia, while videoblogs, tagged images, and media resources that fosters an environment of significant P2P-based video-sharing networks are three exam- individual participation and that promotes community ples of social multimedia. curation, discussion and re-use of content.” We can expand t Social interaction captured in multimedia. Here, the scope of this basic definition by including three types multimedia tools or data capture social activities of interaction: and interactions, a very important aspect of social multimedia due to the proliferation of surveillance t Content interaction between multimedia. Any multime- cameras. Video surveillance offers a moment-by-mo- dia application is composed of more than one medium, ment picture of interactions over extended periods, with different media correlated but not necessarily providing information about both the structure and time-based or colocated. For example, a user might content of relationships. 1 put together an image and several descriptive keyword tags on Flickr to evoke a particular viewer response. While its scope is still expanding, our definition of t Social interaction around multimedia. Here, the em- social multimedia can be expressed as “multimedia re- phasis is on the relations among people within a sources and applications designed to be disseminated

28 COMPUTER

through social interaction or be used to capture social activity and Computational Social multimedia Multimedia interaction.” social science computing technology

SOCIAL MULTIMEDIA Technological infrastructure COMPUTING: Communication Social network Signal Computer Information WHAT IS IT? and HCI theories analysis processing vision retrieval Social multimedia poses a significant challenge for both social Social Organization Visual/auditory Cognitive Computing Sociology Anthropology computing and multimedia re- psychology theory physiology psychology theory search. Social computing raises Theoretical underpinnings many issues, such as modeling the interaction between multimedia and analyzing captured social Figure 2. A cross-disciplinary research and application field. The theoretical underpinnings of social multimedia computing include both computational and social interaction and activity, which is sciences. often beyond the general field’s scope. The fundamental challenge in the multimedia domain addresses the semantic gap be- helps infer image semantics (to annotate the image as a tween the low-level features that we can extract from data banquet, party, or meeting, for example). Here, collabora- and the semantic interpretation that the same data have tive tagging is a process in which users add and share tags for a user in a given situation—and it still remains open. for photos, audios, or videos. Figure 1 shows two views of social multimedia comput- This contextualized paradigm incorporates and ex- ing. The first is social computing over multimedia, which ploits social constraints and contextual information to centers on social sciences. It focuses on using technology infer the semantics of multimedia content. To make a and tools to enable more powerful social interaction, to further step forward, a variety of “local” semantic un- harvest large-scale digital traces and develop methodolo- derstanding engines could be integrated into a network gies for large-scale validations of social science theories, such that each node learns from each other and benefits and to study user behavior and social dynamics in multi- from individual user actions and local context informa- media social networks. tion.3 However, emerging social multimedia applications This not only expands the research scope of social com- also host a wide spectrum of computing issues, such as puting but also provides new tools for computational social wide-area-threat analysis in video surveillance networks, sciences. Consider the example shown in Figure 1. We collaborative health monitoring with ubiquitous wireless can perform social network analysis based on collabora- sensors, and so on. To address these problems, new com- tively tagged data within a social computing framework. puting models and methodologies could be developed by However, we can also build multimedia social networks closely integrating multimedia technology and computa- based on image analysis by, for example, modeling the tional social science. co-occurrence of individuals in group-meeting photos. As a cross-disciplinary research and application field, This approach offers the additional advantage that it ef- the theoretical underpinnings of social multimedia com- fectively reduces name ambiguity so that one person can puting include both computational and social sciences, be referenced through multiple name variations in dif- as Figure 2 shows. From computational social science’s ferent situations—or share the same name spelling with perspective, it must involve sociology and anthropology other people—by combining face recognition and natural- from psychological and organizational theories yet inte- language processing. grate communication and human-computer interaction The second view is social-empowered multimedia com- theories. From the multimedia perspective, research relies puting. This approach emphasizes applying knowledge on the theoretical and technological findings from visual/ from social studies to design and improve multimedia auditory physiology, cognitive psychology, signal process- applications, including harvesting more accurately la- ing, computer vision, communication, and information beled data and deriving metadata from social activities retrieval, to name a few. and resources, using social network analysis and socially collected data for content understanding, exploiting social MAJOR APPLICATION AREAS dynamics to improve multimedia communication and In a broad sense, we can use social multimedia comput- content protection, and employing user behavior analysis ing in any application area that uses social multimedia as to recommend multimedia resources to users. Consider input, such as online content-sharing sites. It also offers again Figure 1’s example, in which collaborative tagging different avenues for the multimedia domain by improving

AUGUST 2010 29

COVER FEATURE

P2P network consists of nodes, or peers, that act as both existing multimedia applications and spawning attractive resource suppliers and consumers of resources; they can alternatives. share a portion of these resources—such as disk storage and network bandwidth—with other peers. However, the Online content-sharing lack of authority or structure poses several challenges for services and communities unstructured P2P systems,6 including free-riding, the exist- In the history of social multimedia, YouTube is un- ence of firewalls/NATs, security breaches, and malicious doubtedly one of the major milestones, mainly because behavior, including cheating, whitewashing, and collusion. it created a platform that provides an attractive user ex- To alleviate these problems, we can model P2P net- perience around sharing video online. YouTube’s success works as multimedia social networks and then analyze triggered other online content-sharing sites such as Flickr, user behavior and the impact of human dynamics on Digg, and Youku, all of which offer users the option to multimedia communication. Modeling P2P networks upload, share, and tag images, audios, and videos, and as social structures can allow incentive, reputation, or create social networks by designating contacts or friends. payment mechanisms to reward good peers and punish A similar service, videoblogging, or vlogging, combines misbehaviors so that peers are more inclined to cooper- embedded videos or video links with supporting text and ate. Such modeling and analysis provides fundamental images.4 A typical example is Barack Obama’s vlog in the guidelines to better design multimedia networking sys- 2008 US presidential election, www.youtube.com/user/ tems. A recent survey of Skype, for example, showed that the performance problems resulting from free-riding and NATs could be reduced by applying social networks As a surrogate to content-based searches in P2P systems.6 currently in their infancy, collaborative recommendation is an important tool for Social multimedia search finding multimedia content. Multimedia search provides an important application area for social multimedia computing. The proliferation of user-generated content (UGC) and the associated metadata ______BarackObamadotcom. Compared to online video-sharing on social multimedia sites introduces new challenges in services, vlogs demonstrate stronger social characteristics, search,7 including vulnerability to spam and noise, and such as social networks. They also provide better social- short lifespan. Further, much of the content offers little interaction data to facilitate analyzing temporal interaction value to the general public, and access control restricts dynamics because entities are often archived in reverse most UGC messages to only a few recipients. Thus by using chronological order. Online content sharing and vlogging social network analysis and socially collected data, social have experienced tremendous growth in the past several multimedia computing could enable improved content years and created a huge marketing opportunity. Accord- analysis. ing to eMarketer, online content-sharing sites will attract As a surrogate to content-based searches currently in 101 million users in the US and earn $4.3 billion in ad their infancy, collaborative recommendation is an impor- revenue by 2011. tant tool for finding multimedia content. For example, Social multimedia computing, together with online developers estimate that 53 percent of online video search- communities, could help create more capable computa- ers discovered online video content through friends.4 So tional infrastructures to support interaction, group activity, user-behavior models and multimedia social networks and collaborative work. Google’s Picasa offers one exam- could be used to create a recommender system that helps ple. Although originally designed as a software application people find images or videos and potential collaborators. for editing and organizing digital photos, people now use Picasa to collect, share, and tag photos. Another example Interactive services and entertainment is in YouTube’s video response mechanism, which lets Interactive service is one of the most promising appli- users provide reviews for products or places and exchange cation areas of social multimedia computing—one such opinions about certain topics through a much richer media example is online video advertising. Although Web ad- than simple text. This video-based interaction opens new vertising is interactive by nature, hyperlinked videos and doors for originality and spontaneity in user interactions.5 vlogs offer a unique and more complex level of engagement with their precise targeting capability.4 This new Multimedia communication advertising model is less intrusive, displaying advertising Ideas and technologies from social multimedia information only when the user chooses it by clicking on computing have recently found their applications in com- an object in a video. By learning user preferences through munication channels such as peer-to-peer networks. A multimedia social network analysis, the hotspots that cor-

30 COMPUTER

respond to brands could be further Applications highlighted to extract more inter- Online content-sharing Social multimedia Healthcare Security ests from users. services and communities search applications An increasing trend is to har- Relationship discovery Privacy protection ness the wisdom of crowds. This from social multimedia is particularly true in interactive entertainment such as gaming, Traitor-tracing and Multimedia social networks storytelling, and edutainment. Re- copyright protection Imagery: relationship discovery and prediction Gaming-driven: multimedia social dynamics Social multimedia cently, researchers have explored Interaction-driven: fusion analysis of content, community analysis collective intelligence in the form Contextualized media network, and context understanding of online games, called games 8 with a purpose. By playing these Cooperative multimedia Multimedia interaction games, people contribute to their distribution dynamics understanding of entities on the Social multimedia computing Web and even collectively solve large-scale computational prob- Computational social science theory, multimedia, and communication technologies lems such as categorizing online pictures, monitoring distributed Figure 3. Various facets of social multimedia computing in a research field where security cameras, and improving multimedia social networks play a core role. online video search.

Healthcare of domains, such as homes, banks, airports, and conven- According to recent public health findings, physical ience stores. Various government entities have used social health factors such as obesity, emotional health factors network analysis to analyze terrorist networks, communi- such as happiness or depression, and harmful habits such cations, criminal organizations, and resources. However, as smoking can significantly affect an individual’s social to date, few works have focused on social network analysis network. Using online social networks for healthcare pro- in surveillance video. vides an opportunity to analyze behavioral data and study In spite of the possible invasion of privacy, mastering social structures formed as a result of ties to health behav- the role such networks play in monitoring surveillance iors. Both the American Cancer Society (cancer.org) and video data is of great interest to law enforcement and the Centers for Disease Control and Prevention (CDC.gov),______homeland security. For example, DARPA is soliciting infor example, have experimented with virtual communities novative research proposals to develop the Persistent Stare such as Second Life to test whether social multimedia can Exploitation and Analysis System for automatically and help spread the word about such issues as nutrition aware- interactively discovering actionable intelligence through ness, cancer screening, and infectious-disease prevention. wide-area-threat analysis of complex motion imagery sur- Collaboratively monitoring health status across media veillance of urban, suburban, and rural environments. with ubiquitous wireless sensors and cameras could also help medical staff craft an effective healthcare applica- RESEARCH ISSUES tion. For example, an ECG sensor carried by an elderly As mentioned earlier, social multimedia computing in- person with physical disabilities might capture any unu- volves cross-disciplinary research between multimedia sual heartbeat rates and send this information through a technology and computational social science. Indeed, it social network of family members, physicians, friends, and might benefit from the past and ongoing research out- emergency services;9 the information’s context—at home comes in both related areas. However, many challenges to or outdoors, for example—could help physicians make social multimedia computing are uniquely separate from time-critical decisions. However, the usage of such sensor those in other systems. data will also give rise to privacy issues. In practice, it could Figure 3 summarizes the various facets of social mul- be less privacy invasive if these sensor data are strictly timedia computing in a specific field of research in which restricted for use in healthcare applications. multimedia social networks play a core role. Take the multimedia social network shown in Figure 1, for example. Security applications We can construct it from the content of publicly available We can characterize many security applications as photo albums on the Web. If we find at least one photo social multimedia computing applications. They’ve in which two people appear together, we might assume emerged especially rapidly in recent years with the pro- they are friends. But because most pictures in Web photo liferation of video surveillance systems in a wide variety albums are taken in the wild, it’s difficult and challenging

AUGUST 2010 31

COVER FEATURE

Sources

Tags Users Social network S Social data network S N on Flickr N 2 1 S Formed by N 3 collaborative activities P2P social (e.g., social tagging) Colluder social network network Interaction-driven social network Gaming-driven social network Perform interaction, collaboration, or other Try to maximize their own payoffi activities in the online communities by participating in resource sharing and exchanging activities

Internet

For surveillance For healthcare Imagery social network

Captured by photos, surveillance videos, Cyberspace or wireless sensors

Real world

Human social network

Figure 4. Multimedia social networks. An imagery social network captures social relationships and activities between real-world users, sharing them through photos, videos, and sensors.

to recover such friend networks using existing face detec- resources. To achieve effective cooperation, network tion and recognition algorithms.10 members observe and learn how others play the game and adjust their own strategies accordingly. Typical Multimedia social networks examples include P2P social networks6 and colluder A multimedia social network is a social network in social networks.11 P2P social networks provide key which a group of users share and exchange multimedia information for designing more efficient multimedia content, as well as other resources.11 However, a quick networking systems, whereas analyzing user behav- analysis can reveal that this brief definition doesn’t cover iors in colluder social networks helps design more all the applications previously covered. Figure 4 shows our secure multimedia content management systems.11 understanding of multimedia social networks, which we t Interaction-driven social networks characterize rela- identify as at least three different networks: tionships based on users’ interaction, collaboration, and other activities in online communities. These t Imagery social networks capture the social relation- kinds of networks provide a basis for the development ships and activities between users through photos, of online social multimedia services such as online surveillance videos, or wireless sensors. Such net- video advertising. works provide an effective way to analyze human social networks in the real world, independent of In addition, a social data network can be obtained from humans’ active role in sharing and exchanging data. online activities such as tagging and collaborative recom- However, this may also lead to privacy problems be- mendation. This isn’t exactly a social network between cause it’s possible to infer the user’s location or other users, but it’s easy to derive on the basis of data. Social private information from these data. data networks can help improve multimedia understand- t Gaming-driven social networks let users maximize ing and analyze community dynamics around multimedia their own payoff by exchanging and sharing their resources such as modeling the structure and evolution of

32 COMPUTER

the community, and analyzing the information diffusion driven social networks and social data networks. In mechanism through social multimedia. general, three types of social data and knowledge can be used in social multimedia computing: Key challenges Although work in social network analysis, multimedia t data such as user behaviors, preferences, or interests; content analysis, and other disciplines can be adapted to t social context such as social network, structure, and social multimedia computing systems, numerous prob- relationships; and lems in modeling, analyzing, and utilizing multimedia t socially collected data such as tags. social networks have yet to be solved. Multimedia social dynamics. Social networks often To make use of these data types, we must discard the involve many users of different types—from rational to assumption of independent data instances that under- selfish to malicious—all with different objectives.11 Thus, lies most statistical-learning-based multimedia analysis modeling and analyzing user behaviors and social dynam- systems. This naturally motivates a collective inference ics poses a fundamental challenge to help stimulate user paradigm that makes simultaneous statistical judgments cooperation, maximize overall system performance, and about the same variables for a set of related data instances. minimize the damage caused by malicious users. Another important challenge relates to quality, espe- Several specific multimedia properties make analyzing cially for socially collected data such as tags. Given the multimedia social dynamics different from traditional uncontrolled nature of collaborative tagging, the diversity social network analysis. First, user behaviors are highly dy- of knowledge such as polysemy or synonymy, and the cul- namic, especially when users watch live streaming video tural background of various users, two prominent issues on the same wireless network or share the same limited dominate social tagging systems: inconsistency and ambi- backbone connection to the Internet. Second, the modeling and analysis of user behaviors are mostly content-relevant. In colluder social networks, To make use of these data types, for example, multimedia fingerprinting or other content we must discard the assumption identification technologies could model user behavior and of independent data instances that track people who illegally use copyrighted multimedia.11 underlies most statistical-learning- Third, the potential rewards are time-sensitive. For exam- based multimedia analysis systems. ple, the earlier a colluded copy is released, the more people will be willing to pay for it. Thus, all colluders have an incentive to mount collusion as soon as possible. guity. Although some studies focus on denoising tag sets Relationship discovery and prediction. Relationship and improving tag consistency in social multimedia data- discovery and prediction are the basic computational sets, the tag quality improvement problem remains open. problems underlying social networks. They become more Further, several content-sharing sites such as YouTube and challenging in multimedia social networks because they Viddler have recently begun offering users the ability to tag must also include the social interactions around and cap- video clips temporally. Unlike tags that describe the entire tured in multimedia. video clip, temporal tags can indicate the user’s feelings In imagery social networks, for example, descriptive toward the current scene.12 information—such as a person’s visual appearance and the actions or activities between people—must be extracted Contextualized media understanding to understand participation in an event or a relationship Although researchers have achieved some interesting re- in some other network. In this case, the relationship dis- sults in the semantic understanding of multimedia content, covery and prediction problem seeks to determine the using content-based analysis technologies to understand extent to which the relationship and its evolution can be specific media, such as discerning a sunset from a sunrise modeled via features. This is exaggerated by low-quality in a picture, is still difficult. Tagging might not solve this images taken under poor conditions and further hampered problem, but it does bring a new perspective of contextual by the complex background of surveillance videos or the and social understanding to multimedia content.3 imperfections of existing object detection and recognition For contextualized media understanding, we propose techniques. three levels of meaning. First, the social context of net- Fusion analysis of content, network, and context. works, relationships, and socially collected data such as Technologically, the analysis of multimedia social net- tags should be exploited in multimedia systems. This work works must be closely integrated with multimedia content has attracted increasing research attention in recent years. analysis and incorporate contextual information as much Experimental results have also verified its effectiveness in as possible. This is especially important for interaction- improving the performance of multimedia retrieval.

AUGUST 2010 33

COVER FEATURE

Second, the local context of users, such as cultural Copyright protection constraints, should be taken into account. Clearly, dif- Copyright protection isn’t a new issue in the multime- ferent users with different cultural backgrounds might dia community, but it’s exacerbated by the wide spread have a diversified understanding of the same image or of social multimedia on the Internet. Obviously, to com- video. Thus, the specialized solutions that exploit local pletely solve this issue requires social, legislative, and cultural constraints might lead to personalized products technical efforts. and services. To address copyright protection, digital rights man- Third, a variety of “local” semantic understanding en- agement has been widely employed. Encryption and gines could be integrated into a network characterized by watermarking offer two major DRM approaches used in distributed and collaborative intelligence. Local semantic the past two decades, either by proactively encrypting inference mechanisms taking place at various nodes of multimedia content to prevent unauthorized access or such a network can learn from each other and benefit proactively embedding watermarks or digital fingerprints from individual user actions and available local context into the host signal for posterior authentication. information to improve inference quality across all nodes. From the social science perspective, the analysis of This network-level MIR approach, originally proposed by piracy and collusion dynamics on multimedia social net- Alan Hanjalic,3 might dramatically shift the computing works helps develop DRM technologies and introduce paradigm of multimedia understanding and retrieval in related legislative treaties. Pursuing this idea, researchers the years to come. have investigated the modeling and analysis of human dynamics in colluders’ social networks, such as an illegal- user-tracing multimedia forensics framework to exploit Scalability in multimedia coding will fingerprints that trace culprits who use copies illegally.11 play a key role in universally accessing However, this framework might be unhelpful in resolv- high-definition multimedia content. ing the copyright issues if this vast amount of social multimedia content bears no encryption, watermarks, or fingerprints. Thus, researchers must further explore behavior modeling and forensics for multimedia social Cooperative multimedia networking networks and beyond. The idea of modeling multimedia social networks to enhance multimedia networks has been extensively dis- Relationship discovery cussed in the framework of both P2P and mobile ad hoc from social multimedia networking. However, integrating multimedia social net- Relationship discovery research involves several com- works with various coding and transmission technologies putational problems underlying various social multimedia to address the specific properties of multimedia content applications, such as friend network construction from has yet to receive much attention. Web albums, wide-area-threat analysis for surveillance With the advance of multimedia network technologies video networks, group discovery in collaborative recom- and the increasingly fierce market competition, more high- mendation, summarization of social multimedia activities, definition videos, such as 3D and multiview videos, are and so on. being placed online and distributed over heterogeneous Essentially, the problems can be boiled down to three wired, wireless, or mobile networks. Limited bandwidth multimedia data-mining tasks: social network construc- thus becomes an even more critical resource between tion and evolution, missing link inference and hidden users of multimedia social networks. relation discovery, and group discovery. Similar topics In this case, the scalability in multimedia coding have been widely explored in data mining and social net- will play a key role in universally accessing high- work analysis, but they represent completely new and definition multimedia content. In general, a scalable challenging tasks in multimedia social networks because coding scheme encodes a high-quality video into sev- social activities constantly change with the mishmash of eral bitstream layers of different priorities. A receiver, interrelated users and media objects. depending on its capability, can subscribe to the base Moreover, most of these tasks depend greatly on the layer with just the basic playback quality or subscribe preprocessing of multimedia content such as feature ex- to additional layers that progressively refine the re- traction, object detection, recognition, and tracking, which construction quality.4 Therefore, the scalable coding are widely recognized as difficult problems in computer scheme, together with the requirements of large-scale vision and multimedia analysis. In this case, investigat- real-time streaming, makes modeling and analysis of ing accurate probabilistic inferences that allow pooling social dynamics in multimedia social networks much multiple, weak pieces of evidence to improve overall per- more complex. formance might be highly effective.

34 COMPUTER

Multimedia interaction dynamics address—or ancillary data such as the IP address or time of Intuitively, developing new multimedia social interaction connection. Properly managing privacy issues is essential tools requires a close integration of sociology, multimedia, both to facilitate research and safeguard consumer pri- and communications technologies. Specifically, the social vacy.1 Nevertheless, the privacy issues surrounding social science studies addressing the dynamics of large-scale multimedia data are more complex. Generally speaking, social interaction and activity might revitalize research users should have the right to control their personal data, into novel social interaction methods and tools. Twitter, which implies getting access to the data, modifying it, originally conceived as a mobile status update service that asking for corrections, or asking for deletion. However, provides an easy way to keep in touch with friends, offers this right isn’t easily guaranteed in the context of social one example. Twitter users send and receive short, frequent multimedia. For example, users are usually eager to answers to one question, “What are you doing?” share pictures in online communities, but these images However, Twitter changed that question to “What’s hap- can easily be used for secondary purposes such as face pening?” so that people, organizations, and businesses could recognition and image retrieval, especially when tagged leverage the network’s open nature to share anything they with metadata such as name, e-mail address, and physical wanted, including pictures and video. This shift created a address of the person pictured. Because a single dramatic new kind of information network in its users’ social space. incident involving a breach of privacy could produce rules In this sense, social scientists should take a more active role and statutes that stifle the nascent field of social multi- in coping with the challenge of developing new multimedia media computing, a systematic study of privacy issues tools to enable more powerful social interaction. and their corresponding technological, procedural, and A greater challenge requires developing user interfaces rule-related developments must be undertaken to reduce and interaction paradigms to allow seamless communica- security risks and preserve research potential.1 tion and interaction with remote and virtual environments. In practice, this vision is achievable by combining new sensors that cover touch, smell, taste, and motion; immersive he explosive growth of social multimedia on output devices such as large displays; and 3D technology. the Internet is revolutionizing the way content distribution and social interaction work, Social multimedia community analysis while presenting an evolving multidisciplinary Social multimedia provides several effective ways to T research and application field. At present, this harvest the large-scale digital traces of social behaviors, research addresses the descriptive analysis level, but the such as online content sharing, vlogging, and video surveil- potential for developing social multimedia computing lance. With the increasing availability of such pervasive theories and methods remains promising. data, key research challenges will involve developing methodologies for large-scale validations of social sci- Acknowledgment ence theories and for new theories and inferential analysis This work is supported by grants from the Chinese National methods that can analyze this kind of new data. Natural Science Foundation under contract number 60973055 For example, several studies have focused on social and number 90820003. behavior and organization dynamics in online text-blog communities. Without considering the multimedia prop- References erties of the community structure, these social network 1. D. Lazer et al., “Computational Social Science,” Science, analysis models can be directly used in vlog communities. vol. 323, 6 Feb. 2009, pp. 721-723. However, the problem becomes much more complex if 2. F.-Y. Wang et al., “Social Computing: From Social Informat- we consider the interrelation among the content, social, ics to Social Intelligence,” IEEE Intelligent Systems, vol. 22, and temporal dimensions of vlogs. For example, large- no. 2, 2007, pp. 79-83. 3. S. Boll et al., “Contextual and Social Media Understanding scale experiments are needed to verify whether the and Usage,” Dagstuhl Seminar Proc. 08251, 2009; ____http:// “six-degrees-of-separation principle” remains valid in vlog drops.dagstuhl.de/opus/volltexte/2009/2025.______communities, given that the change of media forms from 4. W. Gao et al., “Vlogging: A Survey of Video Blogging Tech- text-only to video might speed up information propagation nology on the Web,” ACM Computing Survey, vol. 42, no. and consequently shorten networks. This study could shed 4, 2010, pp. 1-57. new insights into real-world applications such as online 5. F. Benevenuto et al., “Video Interactions in Online Video advertising and viral marketing. Social Networks,” ACM Trans. Multimedia Computing, Com- munications and Applications, vol. 5, no. 4, 2009, pp. 1-25. 6. J. Altmann and Z.B. Bedane, “A P2P File Sharing Network Privacy protection Topology Formation Algorithm Based on Social Network Most social multimedia data are proprietary, such as Information,” Proc. 28th IEEE Int’l Conf. Computer Com- user profile data—name, place, date of birth, and e-mail munications Workshops, IEEE Press, 2009, pp. 242-247.

AUGUST 2010 35

COVER FEATURE

7. J. Cho and A. Tomkins, “Social Media and Search,” IEEE Jaideep Srivastava is a professor in the Department of Internet Computing, vol. 11, no. 6, 2007, pp. 13-15. Computer Science and Engineering at the University of 8. L. von Ahn, “Games with a Purpose,” Computer, vol. 39, Minnesota. He has established and led a laboratory that no. 6, 2006, pp. 96-98. conducts research in databases, multimedia systems, and 9. A. Rahman, A. El Saddik, and W. Gueaieb, “SenseFace: A data mining. Srivastava is a fellow of IEEE. Contact him at Sensor Network Overlay for Social Networks,” Proc. IEEE [email protected].______Instrumentation and Measurement Technology Conf., IEEE Press, 2009, pp. 1031-1036. Tiejun Huang is a professor in the School of Electronic En- 10. D. Luo and H. Huang, “Link Prediction of Multimedia gineering and Computer Science, Peking University, China. Social Network via Unsupervised Face Recognition,” Proc. His research interests include digital libraries, video coding, ACM Multimedia, ACM Press, 2009, pp. 805-808. understanding, and copyright protection. He is a member 1 1. H.V. Zhao, W. Lin, and K.J.R. Liu, “Behavior Modeling and of IEEE and the ACM. Contact him ______at [email protected]. Forensics for Multimedia Social Networks: A Case Study in Noshir Contractor is the Jane S. & William J. White Pro- Multimedia Fingerprinting,” IEEE Signal Processing Maga- fessor of Behavioral Sciences in the School of Engineering, zine, vol. 26 no. 1, 2009, pp. 118-139. School of Communication and the Kellogg School of Man- 12. S.J. Davis, I.S. Burnett, and C.H. Ritz, “Using Social Net- agement at Northwestern University. His research focuses working and Collections to Enable Video Semantics Acquisition,” IEEE MultiMedia, vol. 16, no. 4, 2009, pp. on factors that lead to the formation, maintenance, and 52-61. dissolution of dynamically linked social and knowledge networks in communities. Contact him at nosh@north-______western.edu. Yonghong Tian is an associate professor in the School ______of Electronic Engineering and Computer Science, Peking University, China. His research interests include machine learning, multimedia content analysis, retrieval, and copyright management. He is a senior member of IEEE. Contact Selected CS articles and columns are available him at [email protected]. for free at http://ComputingNow.computer.org.

Richard E. Merwin Student Scholarship Computer Society Student Branch Chapter leaders should apply for the Merwin Scholarship. Over a dozen scholarships of up to $2,000 each are available, for a 9-month academic year, starting in October, and paid in two installments. Apply at:

http://www.computer.org/portal/web/studentactivities/merwin______Who is eligible? Graduate students, juniors, and seniors in electrical or computer engineering, computer science, or a well-defined computer related field of engineering who are active members of Computer Society student branch chapters are eligible. For more information, see the above link or send email to:

[email protected]______Apply Now! Application deadline is September 30th!

36 COMPUTER

COVER FEATURE SOCIAL- NETWORKS CONNECT SERVICES

Moo Nam Ko, Gorrell P. Cheek, and Mohamed Shehab, University of North Carolina at Charlotte Ravi Sandhu, University of Texas at San Antonio

New services such as Facebook Platform, in touch with users while they’re on these sites. Moreover, Google Friend Connect, and MySpaceID users can enjoy various applications with content from let third-party sites develop social appli- numerous third-party sites: users access social-networking cations without having to build their own sites, where they maintain their profiles; third-party sites social network. These social-networks con- retrieve these profiles, enrich the content, and return them nect services increase access to and enrich to the social-networking sites for consumption by the user user data in the Social Web, although they and, possibly, friends. For example, Facebook users can share music with friends, create playlists, and get concert also present several security and privacy alerts on their profile page by installing the third-party challenges. music application iLike (www.ilike.com).______Major social-networking sites have begun launching ocial-networking websites let users build social social-networks connect services such as Facebook Plat- connections with family, friends, and cowork- form, Google Friend Connect, and MySpaceID that further ers. Users can also build profiles for storing break down the garden walls of social-networking sites. and sharing various types of content with These SNCSs let third-party sites develop social applica- S others, including photos, videos, and mes- tions and extend their services without having to either sages. Updating user profiles with interesting content is a host or build their own social network. This extension form of self-expression that increases interaction in such allows third-party sites to leverage the social-networking sites. To encourage this interaction and provide richer con- site’s features. tent, social-networking sites expose their networks to Web For example, third-party sites can exploit the authenti- services in the form of online application programming cation services provided by a social-networking site so that interfaces. These APIs allow third-party developers to users need not create another username and password interface with the social-networking site, access informa- to access the third-party site; instead, users can draw on tion and media posted with user profiles, and build social their social-network credentials and established profile. applications that aggregate, process, and create content Users can also access third-party sites that leverage social- based on users’ interests. network user-profile content. The third-party sites retrieve Social-networking sites provide numerous applica- users’ profiles from the social-networking site to create tion services that can mash up user-profile data with an enhanced experience. In this way, they can increase third-party data. In addition, third-party sites can rapidly membership by providing more interesting content from distribute their services via social-networking sites to keep a variety of sources in a seamless manner.

COVER FEATURE

Social-networks User data in the t Authorization governs access to user connect services Social Web data in the Social Web based on pre- Identity Identity defined authorization access rights; the authentication authorization API lets third-party sites Profile create new content and extract existing Authorization content from users’ Social Web data. Third-party Privacy policy site Streams t Streams let third-party sites publish Client Social graph to users’ activity streams and vice versa. Applications Content t Applications let third-party sites develop rich social features in the form of applications and thereby extend the Figure 1. Social-networks connect services framework. Social Web.

The implementation of these APIs can vary widely with different protocols and technologies. Facebook Platform User data in services Facebook Facebook Platform Authentication Identity OAuth 2.0 Facebook Platform became generally available in December 2008 under Authorization Profile OAuth 2.0 the Facebook Connect brand. (Facebook Third-party Privacy policy announced in April 2010 that it’s eliminat- site Open Stream ing the Facebook Connect brand as part Client Social graph of the launch of the Open Graph API; it’s Applications Content not clear at this point whether the com- -") )( -# -% -! pany will replace it or call it something plug -ins else.) Since then, more than 80,000 third- - vaScript -$%& party sites have implemented it. Facebook Platform lets third-party sites integrate with Facebook and send information both ways to create more engaging and Figure 2. Facebook Platform services. richer social experiences on the Web. As Figure 2 shows, Facebook Platform allows SOCIALfiNETWORKS CONNECT SERVICES users to import their identity, profile, privacy policy, social User data is composed of three types of information. graph, and content from Facebook to third-party sites. Identity data describes who I am in the Social Web, includ- Authentication is by far the most used Facebook Plat- ing my identity, profile information, and privacy policy. form component. This API enables third-party sites to Social-graph data represents who I know in the Social Web, leverage Facebook as an identity provider. For example, including my friendship connections with descriptions Digg.com, a social news website where users can share such as family, coworker, colleague, and so on. Content content, doesn’t require new members to register and data represents what I have in the Social Web, including create a profile. Instead, as Figure 3 shows, they can use my messages, photos, videos, and all other data objects their existing Facebook profile to authenticate (steps 1 and created through various Social Web activities. 2); once authenticated, new users can extend their social For social-networking sites to be able to share user Social graph to the third-party site and invite friends—or link to Web data with third-party sites, a secure and reliable SNCS them, if they’re already members—to join them on Digg. framework is required. As Figure 1 shows, this framework com (step 3). consists of a collection of four categories of APIs that allow Facebook Platform leverages OAuth 2.0 for authentica- third-party sites to interface with the social-networking site: tion and authorization. OAuth 2.0 is a simplified, improved version of the Open Authorization standard (http://oauth. t Identity authentication proves users’ identity; users net)__ that lets third-party sites obtain authorization tokens can authenticate using their existing accounts from from Facebook. First, a user of the third-party site au- various identity providers to include the social- thenticates using Facebook as an identity provider. Next, networking site. Facebook issues a token that lets the third-party site

38 COMPUTER

access the user’s basic profile information including name, picture, gender, and Friend List. The third-party site can request extended permissions depending on User authenticates using the specific application Facebook credentials requirements—for example, to access the user’s Wall, the space on the Step 1 Step 2 user’s Facebook profile where friends can post Step 3 messages. The third-party site can also apply Face- Figure 3. Digg.com authentication via Facebook Platform. New users can leverage their exist- book users’ privacy policy ing Facebook profile to authenticate (steps 1 and 2); once authenticated, they can extend their settings. For example, if social graph to the third-party site and invite their friends—or link to them, if they’re already members—to join them on Digg.com. Alice doesn’t allow Bob to access her content on Facebook, then Bob wouldn’t be allowed to access Alice’s used to integrate third-party sites using Facebook Platform content on the third-party site. along with numerous tools for developers to build social The Open Stream API lets third-party sites read and applications. Facebook Markup Language (FBML) is a pro- write to users’ activity streams—for example, short mes- prietary variant of HTML, and Facebook Query Language sages on new events. This API supports multiple-stream (FQL) provides a quick and easy mechanism to query Face- publishing methods as well as the Atom feed standard. book user data without using API methods. Third-party sites can read content from users’ activity streams in addition to publishing to their activity streams. Google Friend Connect For example, Facebook users can share their activities on Released at approximately the same time as Facebook third-party sites with their friends on Facebook through Platform, Google Friend Connect also makes it easy to the News Feed feature. share profile, social-graph, and content data with third- Facebook provides a series of APIs to assist developers party sites (www.google.com/friendconnect). As Figure in creating social applications that interface with third- 4 shows, its decentralized approach to integrating social party sites (http://developers.facebook.com/docs). The and nonsocial websites relies on open standards such as primary API is Open Graph, which lets third-party applica- OpenID (http://openid.net), OAuth, and OpenSocial (www.____ tions read and write content objects—photos, friends, and opensocial.org).______so on—and the connections among them in Facebook’s OpenID lets users authenticate to third-party sites using social graph. The API is simple in that it allows access to credentials issued by a supported OpenID identity pro- content objects via URLs. For example, with proper au- vider such as Google or Yahoo. This API prevents users thorization, http://graph.facebook.com/FBUserID/friends from having to go through the new-member registration will provide access to FBUserID’s friends. Facebook also process. After initial authentication, users can select their provides support for the Representational State Transfer existing profile from a profile provider site like Plaxo. (REST) API. However, future enhancements will mainly They can then import their social graph from a social- focus on the Open Graph API. networking site such as Orkut to share their activities in Facebook’s social plug-ins also enable traditional third- the third-party site. For example, if a user posts a message party sites to expand into the Social Web with minimal on a board in a third-party site, the message is only visible HTML. The Like button lets users share pages and send to friends in the selected social graph. content back to their Facebook profile. The Recommen- OAuth provides granular authorization control of user dations plug-in allows users to suggest or recommend content in the Social Web based on privacy policies. Users content on the third-party site. The Activity Feed plug-in can share content hosted on a third-party site without shows Facebook users what their friends like. Facebook having to provide a username and password to the request- offers various other social plug-ins including Comments, ing site. OAuth issues authorization to a specific site for a Facepile, Login, and Live Stream. specific content object for a specified time. For example, In addition, Facebook has a JavaScript software devel- assume site A houses photos on behalf of a user and site opment kit consisting of classes and methods that can be B requires access to these photos. Site B requests access

AUGUST 2010 39

COVER FEATURE

The OpenSocial API reads and writes to users’ activity streams on Google Google Friend Identity provider Friend Connect sites, enabling users to Connect services Identity share what they’re doing and to see what Authentication their friends are up to. For example, as OpenID Figure 5 shows, a blog site can leverage Profile provider Authorization Google Friend Connect to publish activi- OAuth Profile ties to various OpenSocial sites. Note the Third-party OpenSocial site $ tivity streams options to “Connect with friends from Client OpenSocial social networks on this site” and “Pub- Application container lish my activities to sites I’ve joined and OpenSocial Social graph connected networks.” $ vaScript $ Content Google developed OpenSocial with $eople and friends the support of numerous other social- $ersistence networking sites. The primary objective $ was to create a platform for third-party site developers to create social applications that integrate with various Figure 4. Google Friend Connect services. OpenSocial containers like MySpace, Hi5, LinkedIn, and Orkut. The advantage of from the user, who then authorizes the access request OpenSocial is that social applications can potentially reach by authenticating to site A and authorizing site B’s access more users if they integrate with multiple container sites, request. Site A will then issue an authorization token for and, theoretically, users will have access to more social ap- the photos to site B. Site B never gains access to the user’s plications. OpenSocial’s client-side API leverages standard site A credentials. Web development technologies such as JavaScript, HTML, and Ajax (Asynchronous JavaScript and XML). It offers server-side development via the RESTful data APIs, which expose three primary sets of data: people and friends—who I am and who I know (social graph); my activities; and persistence—the ability to read and write data with your friends (content). Many third-party sites that adopt Google Friend Connect insert client-side HTML/JavaScript ap- plets, or gadgets, into their pages. In this way, the sites can quickly add social features by simply integrating a few snippets of code. Different social gadgets, such as a rating gadget—which lets users rate a movie—easily transform the third-party site from being part of the Web to being part of the Social Web. The downside of this approach is that gadgets can’t access social graphs or content; developers must use JavaScript or RESTful data APIs to overcome this limitation. Figure 5. Google Friend Connect blog site profile page. The OpenSocial API reads and writes to users’ activity streams on Google Friend Connect sites, enabling users to share what they’re doing and to see what their friends are up to. MySpaceID MySpace has adopted Google’s

40 COMPUTER

Table 1. Social-networks connect services comparison.

API category Facebook Platform Google Friend Connect MySpace ID

Authentication/authorization Single service provider with 400 Multiple service providers Single service provider with 113 million users Open standards: OpenID, OAuth million users Open standard: OAuth 2.0 Open standards: OpenID, OAuth Streams Full support Full support Full support Proprietary Open standard: OpenSocial Open standard: OpenSocial Applications Full suite of APIs and tools Full suite of APIs and tools Full suite of APIs and tools Proprietary Open standard: OpenSocial Open standard: OpenSocial

open-standards approach to Social Web development, has adopted OAuth. Standards-based social applications likewise relying on standards such as OpenID, OAuth, should reach more users because there are so many stan- and OpenSocial to share profile, social-graph, and con- dards-based service providers. However, these users are tent data with third-party sites (http://developer.myspace. also spread across various service providers. Servicing 400 com/myspaceid), as Figure 6 shows. MySpaceID lets users million users with one provider, as Facebook does, creates log in to third-party sites with their MySpace credentials significant mass in the overall Social Web. Not surprisingly, as well as to share their profiles and social graphs with third-party sites want access to Facebook’s membership such sites; it also provides support for activity streams. because it represents an extremely large user base. OpenSocial and MySpace’s own RESTful APIs are the primary resources for creating third-party social applications. OPEN RESEARCH CHALLENGES Because MySpace is an OpenSocial container, MySpace The Social Web is growing exponentially due to SNCSs, user data is easily accessible to any third-party site using but with this growth come several challenges, most per- the standard. taining to security and privacy.

SNCS comparison Identity mapping Facebook Platform, Google Friend Connect, and My- Users often have multiple identities scattered across SpaceID enable third-party site developers to integrate various social-networking and third-party sites. For new with the Social Web without having to build their own users of third-party sites, SNCSs eliminate the need for a social network. Using these SNCSs, users can leverage cumbersome registration process. They also enable users their existing identity, profile, social graph, and content on to link an existing account on a third-party site to one on multiple social-networking sites. All three SNCSs provide a social-networking site, thereby providing a seamless integration APIs and tools. Social Web experience. For example, Facebook compares Table 1 highlights the similarities and differences the hash values of a user’s e-mail address at Facebook with among these SNCSs. One major difference is that Google Friend Connect uses decentralized identity, profile, and social-graph providers while Facebook and MySpace leverage their own social-network plat- MySpaceID Connect User data in MySpace form for user data. The advantage of the services OpenSocial container decentralized approach is that users can Authentication Identity select from among the best of all possible OpenID worlds to customize their presence on the Authorization Profile Social Web; on the other hand, users must OAuth Third-party Privacy policy maintain their social data in multiple lo- site OpenSocial cations, which increases administration Client " tivity streams Social graph costs. Using a single-service provider cuts Application Content down on this administrative overhead but OpenSocial also limits users’ capabilities and choices " vaScript to that of the one provider. " In addition, Google and MySpace follow an open-standards approach while Facebook has traditionally taken a Figure 6. MySpaceID services. proprietary approach, though recently it

AUGUST 2010 41

COVER FEATURE

that at a third-party site and, if they match, links them graphs. Michael Sutterer, Olaf Droegehorn, and Klaus together. It also uses this approach to connect a user’s David proposed an ontology of user subprofiles based on Facebook friends with third-party site friends. However, a specific context.3 Their approach is primarily geared this won’t work if a user’s e-mail address on Facebook toward personalized services on mobile communications differs from that on the third-party site. Moreover, if the platforms but could be applied to the Social Web. For ex- user’s friends have different e-mail addresses, linking their ample, a user could maintain a global profile with multiple accounts is almost impossible. subprofiles that are made available depending on the type Mapping friends’ accounts is especially important of site being visited. in protecting user privacy on the Social Web. Unfortu- nately, current identity-mapping methods like that used Common enhanced privacy policy framework by Facebook have many shortcomings. One ideal solution Social-networking and third-party sites have their own is that all users have a global unique identity offered by access-control mechanisms based on privacy policies that an OpenID service provider. However, OpenID isn’t well use different and often incompatible policy language. In known and hasn’t been fully adopted. addition, current SNCS implementations limit the ability to Researchers have investigated attribute-based identity set and enforce a policy on user data. Facebook Platform mapping. However, G. Alan Wang and colleagues revealed extends users’ privacy controls to social applications hosted that incomplete records would significantly increase the on third-party sites, but only if the application developer comparison algorithm’s error rate.1 This is a common limi- fully implements these capabilities. The approach Google tation of many identity-matching techniques that use only Friend Connect and MySpaceID use to manage the privacy personal attributes. Jennifer Xu and colleagues showed of user profiles and content is even more immature. that combining social features with personal features Several researchers have investigated novel access- could improve criminal identity matching.2 However, this control schemes for the Social Web. The Lockr system approach may not be reliable in today’s Social Web as the lets users manage a single social graph with access-control quality of user attributes in profiles is low due to decep- lists; only those who have a social relationship described tion, errors, or missing attributes. in an ACL can access the shared content.4 Mohammad Mapping multiple identities issued by different identity Mannan and Paul C. van Oorschot proposed an access- providers is another issue in SNCSs. Google Friend Connect control model based on existing “circles of trust” in instant provides authentication services using multiple identity messaging networks.5 And San-Tsai Sun, Kirstie Hawkey, providers. If a user logs in to a third-party site using one and Konstantin Beznosov proposed an e-mail-based identity provider and later logs in to the same site using a system in which users leverage their OpenID identities different identity provider (assuming the user has identities and content owners use their contact lists to specify Open- with both), the third-party site will treat this user as two Policy providers’ access policies.6 The problem with these separate identities and the user will have two profiles, two approaches is that access control is based on one e-mail activity streams, two sets of friends, and so on. Given the contact list, IM network, or social-networking site and thus fact that most users easily forget their usernames and pass- isn’t applicable across multiple social graphs. words, adding an additional dimension—requiring them to Applying a generic privacy policy across all social- select their identity provider—will add a further burden. networking and third-party sites presents a major challenge. Just as OAuth has propelled authentication and User data portability authorization, a common enhanced privacy policy Social Web users often maintain multiple profiles and framework based on open standards must be designed social graphs on different social-networking sites. For ex- and deployed. However, building access controls that use ample, a user may have a Facebook account to keep in compatible policy language and deliver a consistent and touch with family and friends and a LinkedIn account for satisfactory user experience will require extensive col- professional contacts. Managing these disparate profiles laboration within the community. and social graphs can be a chore. One possible solution is for users to maintain a global profile and social graph in Cascaded authorization one place. They could then create subprofiles tailored to Developers can combine data or functionality from two a specific audience—for example, family and friends, col- or more third-party sites to provide a new aggregated ser- leagues, or other members of a virtual community built vice. Third-party sites frequently rely on other third-party around a particular interest—and set appropriate privi- sites to provide services, which requires the sharing of user leges on these subprofiles depending on the context. data from one third-party site to another. Under current There are challenges with this approach: identifying authentication and authorization approaches, users must a single profile repository, resolving duplicate profile at- consent each time a third-party site accesses their data. tributes, and name-space mapping across different social This approach is both cumbersome and time-consuming.

42 COMPUTER

What’s needed is a new authorization mechanism that 4. A. Tootoonchian et al., “Lockr: Social Access Control would obtain a user’s consent once for a specific content for Web 2.0,” Proc. 1st Workshop Online Social Networks object and cascade it to all third-party sites that use that (WOSP 08), ACM Press, 2008, pp. 43-48. content object as part of a mashup service. 5. M. Mannan and P.C. van Oorschot, “Privacy-Enhanced Sharing of Personal Content on the Web,” Proc. 17th Int’l The state of the art is OAuth, which enables third-party Conf. World Wide Web (WWW 08), ACM Press, 2008, pp. sites to share tokens to delegate access rights to other 487-496. third-party sites. The challenge raised by this approach 6. S.-T. Sun, K. Hawkey, and K. Beznosov, “Secure Web 2.0 is that it requires users to trust third-party sites to make Content Sharing beyond Walled Gardens,” Proc. 2009 Ann. authorization decisions on their behalf. OAuth for Recur- Computer Security Applications Conf. (ACSAC 09), IEEE CS sive Delegation is a proposed extension of OAuth that lets Press, 2009, pp. 409-418. 7. B. Vrancken and Z. Zeltsan, “Using OAuth for Re- a client (or third-party site) delegate the authorization it cursive Delegation,” v01, Internet Eng. Task 7 has received from a user to another client. However, this is Force draft, Feb. 2010; http://tools.ietf.org/html/ an emerging standard with no concrete implementations. draft-vrancken-oauth-redelegation-01.______

Data integrity in social plug-ins Moo Nam Kofiis a PhD student in the Department of Soft- ware and Information Systems,fiCollegefioffiComputingfiand SNCSs let people share not only site URLs, events, and Informatics,fiUniversityfioffiNorth CarolinafiatfiCharlotte. His photos with others but their opinions as well. For exam- research interests includefiuser-centric identity and privacy ple, users can share comments about a video they posted management, secure content-sharing frameworks, and on YouTube with friends on a social-networking site. social-application development and methodologies.fiHe is Facebook’s social plug-ins are widely deployed. Several a graduate student member offiIEEE and the ACM. Contact third-party sites have adopted the thumbs-up Like button to him [email protected].______enable users to easily post their opinions on those sites. The Gorrell P. Cheek is a PhD student in the Department of “liked” Web object is referenced and shared using the object Software and Information Systems, College of Computing URL, which makes it easy for a user’s friends to access the and Informatics, University of North Carolina at Charlotte. shared object. However, although social plug-ins enable the His research interests include information security and privacy, with a focus on access-control methodologies. He seamless sharing of user opinions, they don’t guarantee is currently researching machine-learning techniques for data integrity. For example, what happens if the content of governing access within large online social networks. Cheek a “liked” object changes? Adequate data integrity controls is a member of the IEEE Computer Society and the ACM. for social plug-ins remain an open research challenge. Contact him at [email protected]. Mohamed Shehab is an assistant professor in the Depart- ment of Software and Information Systems, College of any third-party sites have adopted social- Computing and Informatics, University of North Carolina at networks connect services to extend their Charlotte. His research interests lie in network and informa- presence in the Social Web. Integrating tion security, especially in the design and implementation these third-party sites with SNCSs creates of distributed access-control protocols to cope with the M a more feature-rich online social com- requirements of emerging distributed social networks, Web munity and promises to break down the garden walls of services, and peer-to-peer environments. Shehab received a PhD in computer engineering from Purdue University. He social-networking sites. However, many challenges come is a member of the IEEE Computer Society and the ACM. with this growth, and the social-networking community Contact him at [email protected]. must collaborate to design and deploy secure services Ravi Sandhu is founder and executive director of the Insti- that both protect privacy and deliver a satisfactory user tute for Cyber Security, holds the Lutcher Brown Endowed experience. Chair in Cyber Security, and is a professor in the Computer References Science Department of the College of Science, with joint appointments in the College of Business and College of En- 1. G.A. Wang et al., “Automatically Detecting Criminal Iden- gineering, at the University of Texas at San Antonio. He is tity Deception: An Adaptive Detection Algorithm,” IEEE Trans. Systems, Man, and Cybernetics, Part A: Systems and also cofounder and chief scientist of TriCipher, an identity Humans, vol. 36, no. 5, 2006, pp. 988-999. solution provider for businesses based in Los Gatos, Calif. 2. J. Xu et al., “Complex Problem Solving: Identity Matching His research focuses on high-impact research, practice, Based on Social Contextual Information,” J. Assoc. Infor- and education in cyber security. Sandhu received a PhD in mation Systems, vol. 8, no. 10, 2007, pp. 525-545. computer science from Rutgers University. He is a Fellow of 3. M. Sutterer, O. Droegehorn, and K. David, “Making a Case IEEE, the ACM, and the AAAS. Contact him at ______ravi.sandhu@ for Situation-Dependent User Profiles in Context-Aware utsa.edu._____ Environments,” Proc. 2007 Workshop Middleware for Next- Generation Converged Networks and Applications (MNCNA Selected CS articles and columns are available 07), ACM Press, 2007, pp. 1-6. for free at http://ComputingNow.computer.org.

AUGUST 2010 43

______

COVER FEATURE A STUDY OF THE HUMAN FLESH SEARCH ENGINE: CROWD-POWERED EXPANSION OF ONLINE KNOWLEDGE

Fei-Yue Wang, Chinese Academy of Sciences Daniel Zeng, Chinese Academy of Sciences and University of Arizona James A. Hendler, Rensselaer Polytechnic Institute Qingpeng Zhang, Zhuo Feng, and Yanqing Gao, University of Arizona Hui Wang, National University of Defense Technology Guanpi Lai, University of Arizona

This first comprehensive empirical study “people-powered” search. A precise definition in a compu- of a search function that originated in tational context is still forthcoming since, as a real-world China examines its tremendous growth in Web phenomenon, HFS is still evolving. Many erroneous recent years and its uniquely rich online/ notions have appeared on various blogs, on wiki sites, and in media reporting, but because they come from non-Chinese offline interactions. sources, they tend to be narrow and often overly specific. For example, a blog entry on SearchEngineWatch.com de- he concept of human flesh search (HFS) origi- fines HFS as “finding and punishing people who publish nated in China where it has become an explosive material Web users consider inappropriate” (http://blog. Web phenomenon in the past five years. In searchenginewatch.com/080627-115435). A June 2008 Times China, Web users routinely employ HFS to iden- Online article (http://technology.timesonline.co.uk/tol/news/ T tify corrupt government officials and individuals tech_and_web/article4213681.ece) characterizes HFS activi- engaged in other illegal or unethical activities. HFS has also ties as “digital witch hunts.” Finally, a guardian.co.uk blog played a positive role; providing public services such as entry (www.guardian.co.uk/commentisfree/2008/nov/02/ helping people find missing relatives during crises. Increas- chinathemedia-blogging)______refers to HFS as “an internet mob ingly, companies and celebrity hopefuls have also started to that hunts down real people online, then verbally abuses exploit HFS as an advertising and public relations platform. them and publishes the victim’s private information.” So, what is human flesh search? The term derives from Chinese-based sources offer broader definitions. For a literal translation of its original Chinese root ெ⫏ᦇ⣬ , example, ChinaSupertrends.com defines HFS as “online which refers to searches that are conducted with help from crowds gathering via China’s bulletin board systems, chat human users (as opposed to on a purely automated plat- rooms, and instant messaging to collaborate on a common form, such as Google), often targeted at finding the identity task” (www.chinasupertrends.com/chinas-human-flesh-______of a human being. A more accurate translation would be search-engine-not-what-you-might-think-it-is).______

COVER FEATURE

To study HFS empirically, we’ve collected a compre- photographed the tiger, and the image subsequently ap- hensive set of online episodes commonly labeled as HFS peared in an issue of Science magazine (“Rare-Tiger Photo events, from their inception in 2001 to 5 May 2010. Our Flap Makes Fur Fly in China,” www.sciencemag.org/content/ data collection strategy involved both automatic Web vol318/issue5852/r-samples.dtl).______crawling and manual search. Sources covered included After these photographs were published in various out- media reports (both online and printed), online forums, lets and posted online, Web users around the world sought blogs, video sharing sites, game chat rooms, and results to prove or disprove their authenticity, leveraging domain from general Web searches, among others. Our dataset expertise that ranged from zoology, botany, and photogra- contains 404 HFS episodes. For all of these episodes, we phy to geometry. Finally, an HFS participant successfully have episode-level information covering items such as the identified the original calendar cover painting from which start and end times of HFS activities, the major types of the hunter had copied the photograph to create and edit online and offline activities that have taken place (often the claimed South China Tiger pictures. Consequently, the reconstructed from secondhand material such as media HFS process ended with proof that the South China Tiger reporting), the nature of the event, and the final outcome photos and claims were indeed scams. of HFS. For 211 of these episodes, we were able to collect Another HFS episode, the “outrageously priced hair- raw postings and discussion threads from 783 URLs on cut” episode, offers an alternative case of a much more 103 distinct websites. The total number of postings in our local nature, in Zhengzhou City, Henan Province, China, in dataset exceeds 1.1 million. 2008. Two students were forced to pay an extraordinary Based on our data, two defining characteristics of HFS amount of money (200 times the advertised price) for a stand out. First, most HFS episodes involve strong offline regular haircut in a local barbershop. The staff working at elements, in the form of either information acquisition the barbershop claimed that they had “deep connections” through offline channels or other types of offline activism. (presumably with the local government and the police) and Second, almost all HFS events rely on voluntary crowd- they were not afraid of any media investigation. The local sourcing: a team of Web users join each other to share people were enraged and started HFS to identify who these information, conduct investigations, and perform other claimed “deep connections” were. A local group was formed actions concerning people or events of common interest. to conduct physical investigations in offline settings as well. From a systems standpoint, HFS engines (HFSEs) are plat- In addition, net citizens demonstrated in front of this bar- forms—dedicated websites and online forums—that enable bershop and vandalized it. The government investigated HFS activities by broadcasting requests and action plans, the case and fined the barber owner a half million RMB sharing online and offline search results, and sometimes, (Chinese Yuan, about $74,000) for illegal pricing schemes. although rarely, offering specific rewards. In our analysis, In recent years, HFS has become an international Web information from offline sources refers to information acces- phenomenon. From 2001 to the end of 2007, about 35 sible (either readily available or acquired through some effort) HFS episodes occurred, with 31 taking place in China (in- to some HFS participants but not yet available from any cluding the Chinese mainland and Taiwan), one in Japan, online sources. Offline actions or activities are those taken one in Korea, and two in the US. In 2008, a total of 114 in the physical world and outside the Web environment. HFS episodes took place, with 111 occurring in China, one in the UK, one in France, and one in the US. In 2009 a EMPIRICAL ANALYSIS total of 168 HFS episodes were reported, with two in the We can trace back the first HFS episode to 2001, when a US, one in Japan, one in Korea, one in Lithuania, and the user posted a photo of a young woman on a Chinese online rest in China. From January 2010 to the beginning of May forum, part of Mop.com, an interactive entertainment web- 2010 there were 87 HFS episodes, with one in the US, one site, and claimed she was his girlfriend. A number of users in the UK, and the rest in China. From the beginning of then performed a range of online and offline searches, 2008 forward, the frequency of HFS episodes has risen seeking to reveal her identity. Finally, one of them posi- drastically, an upward trend that continued throughout tively identified the young woman (who happened to be 2009 and 2010. Figure 1 plots the number of HFS epi- a minor celebrity) and posted her personal relationship sodes quarterly over time. The peak in 2008 is attributed information online, discrediting the claim made by the to activities related to the Beijing Olympics and Sichuan user who initially posted the photo. earthquake. Figure 2 plots the ratio of the number of HFS One of the most famous and possibly biggest episodes episodes quarterly over the number of Internet users in as measured by number of participants involved HFS and China. Clearly, the increase in HFS activities isn’t simply the “South China Tiger” event. In 2007, a hunter in Shaanxi due to a growth in Chinese Internet use. Province, China, claimed to have encountered a live wild HFS episodes typically start with the formation of a small South China Tiger, which, as a species, has long been seed community issuing a task with a defined goal. For more considered extinct in natural environments. This hunter than half of HFS episodes, information about the context

46 COMPUTER

and underlying event appeared first on an 70 online forum, which in most cases became 60 the portal of ensuing HFS-related activities 50 for this particular episode. For the others, “trigger” stories from 40 sources such as TV and online news, web- 30 sites such as tudou.com, or print-based 20

newspapers appeared first on an online Number of HFS episodes forum. Members then started discussions 10 about these stories, which in turn initi- ated HFS activities. Less than 5 percent 0

of these HFS events started with a cash 2001 2002 2003 2004 2005 reward, differentiating HFS from Ama- zon’s “Mechanical Turk” (www.mturk. 2010 1st quarter 2009 1st quarter 2008 1st quarter 2007 1st quarter 2006 1st quarter 2009 3rd quarter 2008 3rd quarter 2007 3rd quarter 2006 3rd quarter 2009 4th quarter 2008 4th quarter 2007 4th quarter 2006 4th quarter 2009 2nd quarter 2008 2nd quarter 2007 2nd quarter 2006 2nd quarter ___com) and other sites where people par- Time period ticipate primarily for monetary gain. We also note that in almost all episodes, Figure 1. Evolution of human flesh search episodes quarterly, over time. the participants include large numbers 0.0025 of people with no particular identifiable 0.002 skill base, differentiating this from open 0.0015 source development activities and other 0.001 skill-based Web communities. 0.0005 As HFS activities move into full swing, 0 participants typically engage in a wide Number of HFS episodes per 2001 2002 2003 2004 2005 range of online and offline activities. In 1 million Chinese Internet users about 50 percent of HFS episodes, Web 2007 1st quarter 2006 1st quarter 2009 1st quarter 2008 1st quarter 2009 3rd quarter 2008 3rd quarter 2007 3rd quarter 2006 3rd quarter 2009 4th quarter 2008 4th quarter 2007 4th quarter 2006 4th quarter 2009 2nd quarter 2008 2nd quarter 2007 2nd quarter users reported their offline findings back 2006 2nd quarter to the online discussion; in about 30 per- Time period cent of them, traditional media or Web Figure 2. Quarterly ratio for the number of HFS episodes over the number of media were involved. Internet users in China. Offline, some HFS participants offered donations or other activities to help victims. In some cases, enraged Web users made phone calls or Announcing even harassed the suspected culprit and associated family members. In October 2008, one HFS episode resulted in the Cash bounty tragic death of an innocent victim (http://news.sina.com. cn/s/2009-04-17/100817629779.shtml).______The girlfriend of a Investigation-Individual young man broke up with him and moved to a different city. In an attempt to locate her, this young man started an Investigation-Group HFS by claiming that he was dying and wanted to see his Uploading exgirlfriend one last time. Out of sympathy to this “dying” HFS online Offiine man and his last wish, the HFS community mobilized and communities environments Investigation- successfully found the girl’s location. The young man went Media Investigation- Media to meet with her, and after an unsuccessful bid to win her Media back, killed her with a knife. After this tragic and senseless Responses event, along with several other less-serious incidents that involved violations of privacy that led to harassment, in- Collaboration-Group sults, and job-related firings, the HFS community engaged Actions-Individual in a postsearch retrospective analysis to determine if the HFS requestors or originators had a hidden purpose. The Actions-Group community examined both the legality and negative social impact of HFS activities. Figure 3 shows the information and control flow be- Figure 3. Information and control flow for HFS online/offline tween online communities and the offline environment. interactions.

AUGUST 2010 47

COVER FEATURE

Table 1. Types of online/offline interactions in HFS. Table 1 summarizes the various types of online/offline interac- Type of interaction Description tions that manifested in these HFS Announcing Announcing on a blog/online forum event information acquired offiine to episodes. Figure 4 shows the preva- trigger HFS lence of various online and offline Cash-bounty Ofiering cash bounty for information interaction types that occurred Uploading Sharing/uploading information identiffled offiine by HFS contributors in during the HFS interactions. response to discussion threads Responses Responses from subjects (individual or organizations) impacted by HFS as NETWORK ANALYSIS OF a consequence of discussion threads and related media coverage HFS COMMUNITIES Investigation Offiine investigation by individual HFS contributors The research opportunities HFS (individual) offers to social and computational Investigation Offiine investigation by reporters from traditional media scientists are boundless and in- (media) clude social network analysis as Investigation Offiine investigation by groups of HFS contributors organized through well. The case studies we describe (group) various HFS channels (online forums, instant messaging) in this section analyze two different Actions (individual) Offiine actions/responses (demonstrations, phone calls, donations) by networks of HFS participants and individual task contributors seek to shed light on two research Actions (group) Offiine actions/responses by groups of HFS contributors organized questions. First, how do HFS par- through various HFS channels ticipants interact? Second, how do Collaboration Offiine collaboration between groups of HFS contributors and real-world HFS online communities differ from (group) organizations other online communities? Our HFS dataset consists of a Announcing large set of HFS episodes occur- Cash-bounty ring between 2001 and May 2010. Uploading For each episode, we employed au- Investigation-individual Investigation-media tomatic Web crawling methods to Investigation-group extract information about partici- Responses pants and the connections among Actions-individual them. The two case studies are

Online/offline interaction types Actions-individual based on two episodes from this Collaboration-group dataset, selected to represent two 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 very different networks—national Percentage of online/offline interaction episodes and local episodes. Na- tional episodes refer to events with Figure 4. Prevalence of various online and offline interaction types occurring during national or global impact that have HFS interactions.

Table 2. Discussion threads analyzed for the “South China Tiger” event.

Discussion threads Start time Milestones

163 News comments #1 2007-10-12 17:42 The photo of the claimed tiger was publicized (Figure 6a) XITEK Discussion #2 2007-10-13 07:30 Doubts were cast upon the photo (Figure 6b) SINA News comments #1 2007-10-13 10:40 TIANYA Discussion 2007-10-15 18:11 SINA News comments #2 2007-10-24 02:25 Ocial investigation commenced (Figure 6c) MOP Discussion #1 2007-10-31 03:41 MOP Discussion #2 2007-11-01 14:51 MOP Discussion #3 2007-11-03 04:51 163 News comments #2 2007-11-14 02:19 Another investigation started XITEK Discussion #2 2007-11-16 10:30 The original image from a calendar cover painting was found (Figure 6d) SINA News comments #3 2007-11-16 14:43 The original calendar cover was publicized; HFS ended

48 COMPUTER

attracted people from all over the country to participate. Local episodes refer to events that mainly took place locally and engaged a local, typically small, Internet user community. About one-fourth of the episodes are classified as national episodes and the rest as local episodes. In the following, we present the network analysis of the two most typical national and local episodes. In our analysis, each node in a participant network corresponds to a unique user ID, typically associated with one distinct HFS participant. The existence of an edge between two nodes indicates the presence of Web posting citations between these two corresponding participants.

Case study I: ‘South China Tiger’ network The “South China Tiger” episode involved many HFS engines, but five (tianya.cn, xitek.com, mop.com, 163.com, and sina.com.cn) played a critical role. For the purpose of constructing the South China Tiger participant network, we manually identified 11 of the most important discus- Figure 5. South China Tiger HFS participant network. The nodes are represented by colors: tianya.cn is green, xitek.com sion threads from these five HFS engines: one from tianya. is pink, mop.com is red, 163.com is blue, sina.com.cn is dark cn, two from xitek.com, three from mop.com, two from gray, and cross-platform nodes are black. For edge colors, 163.com, and three from sina.com.cn; Table 2 shows a intra-HFS engine links are gray; inter-HFS engine links are summary. In total, this network consisted of 5,612 distinct black; and virtual links are red, denoting information flow nodes and 7,930 distinct edges. Figure 5 shows the entire that didn’t appear with URLs or user IDs; the size of nodes represents the in-degree (with the exception of the across- network, with all discussion threads included, and Figure 6 platform nodes, which are highlighted with a larger size).

(b)

(a)

Figure 6. Growth of the South China Tiger HFS network. (a) Only one thread appeared at first, and then (b) the HFS spread to the professional photography forum xitek.com. (c) The official investigation started and then (d) ended with discovery of the original poster.

AUGUST 2010 49

COVER FEATURE

Table 3. Discussion threads analyzed for the “outrageously priced haircut” event.

Discussion threads Start time Milestones

Dahe discussion #1 2008-4-1 17:42 The news was publicized; “call for HFS” posted; and local Internet users started to conduct offiine searches (Figure 8a) Dahe discussion #2 2008-4-2 23:15 Many HFS offiine fflndings collected Tianya discussion #1 2008-4-3 11:28 Event reported on a national forum (tianya.cn) (Figure 8b) Dahe discussion #3 2008-4-3 11:24 Dahe users organized an offiine demonstration that occured at 18:30 (Figure 8c) Dahe discussion #4 2008-4-4 18:27 Users reported online about offiine actions (the barbershop was under siege) one hour ago Dahe discussion #5 2008-4-5 16:14 The local government fflned the barbershop owner (Figure 8d)

shows how this network evolved across various milestones. We used the Cytoscape toolkit (http://www.cytoscape.org) for network visualization.

Case study II: ‘Outrageously priced haircut’ network The “outrageously priced haircut” episode’s participants lived in Zhengzhou city, and the main HFS engine focused on the local online forums hosted on dahe.com, along with another national HFS engine, tianya.cn, which played a lesser role. Table 3 summarizes the major discussion threads and event milestones. In total, this network had 1,157 distinct nodes and 1,547 distinct edges. Figure 7 shows the participant networks on dahe.com and tianya.cn.______The local discussions on dahe. Figure 7. The outrageously priced haircut HFS participant com were much denser than those on tianya.cn.______This is network, with the dahe.com site in red and the tianya.cn site because dahe.com users were mostly local residents of in blue.

(a) (c)

(b) (d)

Figure 8. Growth of the outrageously priced haircut HFS network. (a) Only one thread appeared at first, and then (b) the HFS was reported on a national forum (tianya.cn). (c) The offline group investigation and demonstration started and then (d) ended with the punishment against the barbershop made by local government.

50 COMPUTER

0 100 10 Real data Real data λin = 1.619 λout = 2.484

10–1 10–1

10–2 10–2 Frequency Frequency

10–3 10–3

100 101 102 103 100 101 102 In-degree Out-degree

100 100 Real data Real data λin = 2.206 λout = 2.625

10–1 10–1 Frequency Frequency 10–2 10–2

10–3 10–3 100 101 102 103 100 101 102 In-degree Out-degree

Figure 9. In- and out-degree distributions and the power-law correlation of the South China Tiger network and outrageously priced haircut network.

Zhengzhou and its surrounding areas, and they collabo- than one discussion forum. Although small in number, rated both online and offline intensely during this episode. these nodes played a pivotal role in transferring and shar- Figure 8 shows how this network evolved across various ing information across discussion forums and often across milestones. HFS engines. We also observed marked differences between HFS NETWORK PROPERTIES engines. For example, in the South China Tiger network, OF HFS COMMUNITIES xitek.com communities tended to be much denser than HFS communities are generally sparse, with a net- those from tianya.cn______and mop.com, indicating a higher work density of less than 0.001. This stems from the large level of sharing and more intense discussion. This observa- number of participants who post only simple, uninforma- tion can be partially explained by noting that xitek.com is tive replies to existing postings. Such users typically don’t a forum for photographers and generally attracts profes- contribute constructive findings or substantive informa- sionals with specific types of expertise, whereas tianya. tion and don’t attract follow-up discussions or postings. We cn and mop.com appeal to general Web users who do not refer to these people as casuals. HFS participant networks alwasys have expertise in photography. are typically highly connected. In the case of the South China Tiger network, for example, 5,612 nodes formed a Degree distribution giant connected component. Degree distribution is a widely used metric to describe HFS communities usually grow quickly at the begin- node connectivity in large, complex networks.1 The degree ning of an episode. Every time a major finding is reported distributions of HFS participant networks can be greatly or milestone reached, a spike of new participants joins affected by the existence of a large number of casual par- in and then quickly evaporates. In many HFS episodes, ticipants and huge hubs due to its forum-discussion nature. multiple engines with their own discussion threads were As shown in Figure 9, these out-degree distribu- actively engaged, but few participants appeared on more tions seem to follow a power-law distribution, whereas

AUGUST 2010 51

COVER FEATURE

Table 4. Connectivity comparison among HFS, Web graph, and blogosphere samples.

Weakly connected Strongly connected Network # of nodes components (max) components (max) Fraction of SCCs in WCCs

HFS-South China Tiger 5,612 5,612 (100%) 242 (4.31%) 4.31% HFS-outrageously priced haircut 1,157 586 (50.65%) 35 (3.03%) 5.97% Web4,8 203,549,046 186,771,290 (91.76%) 56,463,993 (27.74%) 30.23% Blog (source from BlogPulse)4 143,736 107,916 (75.08%) 13,393 (9.32%) 12.41%

in-degrees have flat, long tails. However, in most cases, If we treat the networks as undirected, the average shortest these tails formed from a number of central hubs that cor- paths for both are only 4.420 and 2.399, respectively; the respond to original posters of various discussion threads. diameters are 10 and 4, respectively. These numbers are Such hubs actually performed more like a platform than very small compared to the total number of nodes in these an individual participant. If we discount these hubs, the networks—5,612 and 1,157, respectively. In addition, the in-degree distribution also follows power-law distribution. clustering coefficients of the two networks are 0.057 and The powers of in-degree and out-degree were –1.62 and 0.079, many times larger than the theoretical predictions for –2.48, respectively, for the South China Tiger network, and random networks with the same size—0.00025 and 0.001, –2.21 and –2.63, respectively, for the outrageously priced respectively. These observations indicate that the South haircut network. China Tiger and outrageously priced haircut networks show This means that a few participants generated most of the small-world effect, which has been observed in many the citations and a few participants got cited in these cita- kinds of networks, including the Web, the blogosphere, co- tions. We also found that out-degree dropped off much authorship networks, and many other social networks.4–7 more rapidly than in-degree. Further, there were fewer However, possessing the small-world effect doesn’t nec- nodes that cited many other nodes than the nodes that essarily mean that information can diffuse easily. In both received citations from a large number of other nodes. networks, only 1 percent of the node pairs in the network Comparing HFS network findings with those for blogs, are reachable, many fewer than the 22.11 percent for blogs4 we see that the in- and out-degree distributions of the and 25 percent for the Web.8 This is because HFS networks posting-level blogsphere vary significantly among differ- are highly connected via central hubs and most nodes are ent blog servers. In an analysis with IBM’s internal blogs casual nodes, only generating citations and replies to the between November 2003 and August 2006,2 the in-degree central hub and otherwise contributing none. distribution followed power-law with a power of –1.6, com- Another measure used to determine network connectiv- parable to the South China Tiger network, but smaller than ity is the ratio between the maximum strongly connected the outrageously priced haircut network. The out-degree components (SCCs) and the weakly connected compo- distribution followed power-law with a power of –1.9, nents (WCCs) of the network. Table 4 shows our results. which is much lower than both HFS networks. We found that the connectivity of HFS network samples is Another empirical analysis based on data from differ- weaker than that of the Web and blogosphere. The fraction ent blog sites revealed that for blog networks, the in- and of the max WCC in the South China Tiger network reached out-degree distributions both followed a heavy-tailed dis- 100 percent, a result of parallel searches across different tribution, with a shallow power-law exponent of –1.7 for HFS engines. The outrageously priced haircut network, on the in-degree distribution. The in- and out-degree distribu- the other hand, split into two parts. tions of the post network followed a power law (exponents –2.1 and –2.9 for in- and out-degree distributions).3 Evolutionary patterns Other researchers also reported that the out-degree We conducted several studies to examine the evolution- distribution for blog networks have slight deviations from ary patterns of HFS participant networks and discovered power-law distribution, possibly because of data sampling the following: as previously noted, both the South China limitations.4 In our studies, most of the HFS networks’ out- Tiger network and the outrageously priced haircut network degree data fit the power-law distribution very well. had a large max WCC but a relatively small max SCC. We further observe that for the South China Tiger network, Small-world efiect and connectivity through slicing the entire duration of the episode into The average shortest path is 5.167 for the South China 10-day periods and comparing observations over these Tiger network and 4.331 for the outrageously priced haircut periods, both the size of the max SCC and max WCC— network, with a diameter of 12 for the South China Tiger net- the network itself—grew larger, while the connectivity work and nine for the outrageously priced haircut network. increased with network growth. We also observed that

52 COMPUTER

while the fraction of WCCs (100 percent) didn’t change, the 2. P. Kolari et al., “On the Structure, Properties and Utility size of WCCs continued to grow with the number of nodes; of Internal Corporate blogs,” Proc. Int’l Conf. Weblogs and the max SCCs’ growth slowed after reaching a certain size. Social Media (ICWSM 07), AAAI Press, 2007, pp. 113-120. This indicates that the clustering coefficient decreased 3. J. Leskovec et al., “Patterns of Cascading Behavior in Large Blog Graphs,” Proc. 7th SIAM Int’l Conf. Data Mining (SDM with time, which is the opposite of the blogosphere, while 07), SIAM Press, 2007, pp. 551-556. the fraction of reachable node pairs increased with time. 4. X. Shi et al., “Looking at the Blogosphere Topology through Jure Leskovec and colleagues described the densifica- Different Lenses,” Proc. Int’l Conf. Weblogs and Social Media tion law prevalent in many networks,9 which is also the (ICWSM 07), AAAI Press, 2007, pp. 153-160. case for the South China Tiger and outrageously priced 5. R. Albert et al., “The Diameter of the World Wide Web,” haircut networks: the number of edges grew superlinearly Nature, vol. 401, 1999, pp. 130-131. to the number of nodes over time, with the densification 6. R. Kumar et al., “The Web and Social Networks,” Computer, vol. 35, no. 11, 2002, pp. 32-36. exponent D = 1.21 for the South China Tiger network and 7. M. Newman, “The Structure of Scientific Collaboration D = 1.07 for the outrageously priced haircut network, Networks,” Proc. Nat’l Academy of Sciences, vol. 98, no. 2, comparable to the value for the graph of routers in the In- 2001, p. 404. ternet (1.18), smaller than the values for the patent network 8. A. Broder et al., “Graph Structure in the Web,” Computer (1.66)9 and the blogosphere (1.928).4 Networks, vol. 33, nos. 1-6, 2000, pp. 309-320. 9. J. Leskovec et al., “Graphs over Time: Densification Laws, uman flesh research is an important area for Shrinking Diameters, and Possible Explanations,” Proc. 11th ACM SIGKDD Int’l Conf. Knowledge Discovery in Data cross-disciplinary research, in part because it Mining, ACM Press, 2005, pp. 177-187. is a significant Web application drawing world- 10. E.H. Chi, “Information Seeking Can Be Social,” Computer, wide attention and because it has a strong vol. 42, no. 3, 2009, pp. 42-46. H and unique cultural root. From a technical 11. F.-Y. Wang et al., “Social Computing: From Social Informat- research perspective, HFS shares many common character- ics to Social Intelligence,” IEEE Intelligent Systems, vol. 22, istics with emerging social search engines, yet encourages no. 2, 2007, pp. 79-83. a much broader range of online and offline interactions.10 12. F.-Y. Wang, “Toward a Paradigm Shift in Social Computing: HFS provides a fruitful dataset that can be used to de- The ACP Approach,” IEEE Intelligent Systems, vol. 22, no. 5, 2007, pp. 65-67. velop and test new social networks and complex network theories and models, while also conducting research into network-centric machine learning and data mining.11 Fei-Yue Wang is a professor at the Institute of Automa- We envision that the unique online/offline interactions tion of the Chinese Academy of Sciences. Contact him at abundant in HFS episodes could provide critically needed [email protected]. insights facilitating the development of new social theory Daniel Zeng is a research professor at the Institute of Auto- that blends traditional social theory, which focuses mostly mation of the Chinese Academy of Sciences and a professor on offline behavior, with newer studies focusing on online at the University of Arizona. Contact him at ______zengdaniel@ behavior.11,12 HFS communities can be viewed as a special gmail.com. kind of cyber-enabled social movement organization. The James A. Hendler is a professor of web science at Rensselaer uniquely massive online/offline interactions of HFS will Polytechnic Institute. Contact______him at [email protected]. open new venues for conducting high-impact, data-driven Qingpeng Zhang is a PhD student in the Department of research to empirically test various theories and hypoth- Systems and Industrial Engineering at the University of eses that, until now, have been impossible or prohibitively Arizona. Contact him at [email protected]. expensive to test. Zhuo Feng is a PhD student in the Department of Systems Acknowledgments and Industrial Engineering at the University of Arizona. This work is supported in part by the National Natural Science Contact him at [email protected].______Foundation of China through grants 70890084, 60921061, Yanqing Gao is a research scientist in the Department of and 90924302; by the Ministry of Science and Technology of Systems and Industrial Engineering at the University of China through a 973 grant 2006CB705506 and an 863 grant 2006AA010106; by the Chinese Academy of Sciences through Arizona. Contact her at [email protected]. grant 2F07C01; and by the US Defense Advanced Research Hui Wang is a professor in the College of Information Projects Agency’s Transformational Convergence Technology Systems and Management at the National University of Office through a seedling grant to Rensselaer Polytechnic Defense Technology in Changsha, Hunan, China. Contact Institute. him at [email protected].______References Guanpi Lai is a PhD student in the Department of Systems 1. M. Newman, “The Structure and Function of Complex and Industrial Engineering at the University of Arizona. Networks,” SIAM Rev., vol. 45, 2003, pp. 167-256. Contact him at [email protected].______

AUGUST 2010 53

RESEARCH FEATURE Real-World Distributed Computer with Ibis

Henri E. Bal, Jason Maassen, Rob V. van Nieuwpoort, Niels Drost, Roelof Kemp, Timo van Kessel, Nick Palmer, Gosia Wrzesi ska, Thilo Kielmann, Kees van Reeuwijk, Frank J. Seinstra, Ceriel J.H. Jacobs, and Kees Verstoep Vrije Universiteit, Amsterdam

The use of parallel and distributed computing systems is essential to meet the ever-increasing computational demands of many scientific and industrial applications. Ibis allows easy programming and deployment of compute-intensive distributed applications, even for dynamic, faulty, and heterogeneous environments.

he past two decades have seen tremendous on such distributed systems.3 In addition, research hasn’t progress in the application of high-perfor- adequately addressed the problems that can arise from mance and distributed computer systems in combining multiple unrelated systems to perform a single science and industry. Among the most widely distributed computation. This is a likely scenario, as many T used systems are commodity compute clus- scientific users have access to a wide variety of systems. ters, large-scale grid systems, and, more recently, In itself, each cluster, grid, and cloud provides well- economically driven computational clouds and mobile defined access policies, full connectivity, and middleware systems. In the last few years, researchers have inten- that allows easy access to all its resources. Such systems sively studied such systems with the goal of providing are often largely homogeneous, offering the same software transparent and efficient computing, even on a world- configuration or even the same hardware on every node. wide scale.1 Combining several systems, however, is apt to result in a Unfortunately, current practice shows that this goal re- distributed system that is heterogeneous in software, hard- mains out of reach.2 For example, today’s grid systems are ware, and performance. This may lead to interoperability mostly exploited to run coarse-grained parameter-sweep problems. Communication problems are also probable or master-worker programs. For more complex applica- due to firewalls or network address translation (NAT), tions, grid usage is generally limited to straightforward or simply because the geographic distance between the scheduling systems that select a single site for execution. resources makes efficient communication difficult. More- This is unfortunate, as many scientific and industrial ap- over, a combination of systems is often dynamic and faulty, plications—including astronomy, multimedia, medical as compute resources can be added or removed or even imaging, and biobanking—would benefit from distributed crash at runtime. The use of inherently heterogeneous and compute resources. Optical networking advances also unreliable resources such as desktop grids, stand-alone enable a much larger class of applications to run efficiently machines, and mobile devices exacerbates these issues.

REAL-WORLD DISTRIBUTED COMPUTING Clouds Desktop grids An ad hoc collection of compute resources that communicate with one another via some network connection constitutes a real-world distributed system, as shown in Figure 1. Writing applications Clusters Grids for such systems is notoriously difficult, as application programmers must take into account all of the problems described above. Deploying the applications is equally hard because each site is likely to Mobile devices Stand-alone machines have its own middleware and access policies. The uptake of high-performance distributed computing can be enhanced if these complexities Figure 1. A “worst-case” real-world distributed system consisting are abstracted away by a single software system of clusters, grids, and clouds as well as desktop grids, stand-alone that applies to any real-world distributed system. machines, and mobile devices. Clusters, grids, and clouds are well- organized subsystems that use their own middleware, programming Conceptually, such a system should offer two interfaces, access policies, and protection mechanisms. logically independent subsystems: a programming system, offering functionality traditionally associated with programming languages and communication that offer support for fault tolerance and malleability— libraries; and a deployment system, offering functionality adding and removing machines on the fly—and that associated with operating systems. The programming automatically circumvent any connectivity problems. system should allow applications to be not only effi- The deployment system should allow for easy deploy- cient but also robust by providing programming models ment and management of applications, irrespective of

Figure 2. Ibis system architecture. The white boxes belong to Ibis; the gray ones represent third-party software.

AUGUST 2010 55

RESEARCH FEATURE

RELATED WORK To serve the vast majority of users, ranging from system and application-level software developers to application users, the subsystems should follow a layered approach, bis provides a dual-subsystem solution for real-world dis- with programming interfaces defined at different abstrac- I tributed programming and deployment. In contrast, most tion levels, each tailored to different users’ needs. existing systems focus on one of the two subsystems.1 ProActive,2 like Ibis, follows a logical dual-subsystem approach. IBIS ARCHITECTURE It contains several grid programming models and provides grid deployment and virtualization components. Whereas the Ibis IPL is Researchers are extensively studying high-performance small and highly ečcient, the core of ProActive is more heavy - and distributed computing tools and mechanisms in the weight. ProActive further requires the user to handle all connection field, but no existing single software system covers the full setup problems and select the appropriate middleware. spectrum. Ibis aims to address this deficiency with an in- Phoenix3 consists of a general message-passing model that tegrated system based on a straightforward, user-oriented allows compute nodes to be added to and removed from a running philosophy: real-world distributed applications should be application. It also deals with some of the connection setup prob- developed and compiled on a local workstation, and de- lems solved by SmartSockets. Phoenix provides easy-to-use tools that handle common grid operations. However, it can’t automati- ployed from there onto the distributed system. cally exploit dićerent grid middleware systems simultaneously. This “write-and-go” philosophy requires minimal as- The GRID superscalar framework4 supports a certain degree of sumptions about the execution environment. To enable automatic grid deployment. It consists of an application program- applications to run in a heterogeneous system, Ibis ex- ming interface, a runtime system, and a grid deployment center. It ploits Java virtual machine technology. As Figure 2 shows, automatically converts a sequential application composed of tasks the Ibis system architecture follows the dual-subsystem into a parallel application, allowing independent tasks to be exe- approach. The programming system provides a range cuted on dićerent grid resources. of programming models, all implemented on the same The GridRPC specication5 denes an API for remote procedure communication library: the Ibis Portability Layer (IPL). calls in grids. Two reference implementations exist, Ninf-G and Net- Solve/GridSolve. In contrast to the JavaGAT, the binding of grid The deployment system contains a GUI and a library for middleware to application objects is entirely static. deploying and managing applications, implemented on a The Open Grid Forum is currently standardizing the next-gener- middleware interoperability layer: the Java Grid Applica- ation grid programming toolkit: a Simple API for Grid Applications tion Toolkit (JavaGAT). (SAGA).6 The goal is to provide a simple, uniform, standard pro- Ibis is modular and flexible, allowing users to select gramming interface for distributed applications, with consistent the functionality they require from either Ibis or other semantics and style for dićerent grid functionalities. Notably, software. For example, applications can use the high- SAGA’s Java Reference Implementation is implemented directly on top of the JavaGAT. level programming models and the deployment GUI, but they can also be implemented directly on the IPL and References JavaGAT. Likewise, applications are free to use only one 1. I. Foster and C. Kesselman, eds., The Grid 2: Blueprint of a New of the two subsystems. For example, the deployment Computing Infrastructure, 2nd ed., Morgan Kaufmann, 2003. system can be used to deploy C/MPI and other non-Java 2. L. Baduel et al., “Programming, Deploying, Composing, for the applications. Grid,” Grid Computing: Software Environments and Tools, J.C. Ibis is fully open source and used in various real-world Cunha and O.F. Rana, eds., Springer, 2006, pp. 205-229. 3. K. Taura et al., “Phoenix: A Parallel Programming Model for distributed applications such as multimedia computing, Accommodating Dynamically Joining/Leaving Resources,” spectroscopic data processing (FOM Institute for Atomic Proc. 9th ACM SIGPLAN Symp. Principles and Practice of Parallel and Molecular Physics), human brain-scan analysis (Vrije Programming (PPoPP 03), 2003, ACM Press, pp. 216-229. Universiteit Medical Center), automatic grammar learn- 4. R. Badia et al., “Programming Grid Applications with GRID ing, and many others. In addition, Ibis has been used to Superscalar,” J. Grid Computing, June 2003, pp. 151-170. 5. K. Seymour et al., “Overview of GridRPC: A Remote Procedure build high-level programming systems, including a work- Call API for Grid Computing,” Proc. 3rd Int’l Workshop Grid Com- flow engine for astronomy applications in D-Grid (Max puting, LNCS 2536, Springer, 2002, pp. 274-278. Planck Institute for Astrophysics), the GridChem gate- 6. T. Goodale et al., “SAGA: A Simple API for Grid Applications, way for TeraGrid, and a grid file system (University of High-Level Application Programming on the Grid,” Computa- Erlangen-Nürnberg). tional Methods in Science and Technology, vol. 12, no. 1, 2006, pp. 7-20. Ibis has also been applied to enhance existing systems such as ProActive (INRIA), Jylab (University of Patras), and the GRID superscalar framework (Barcelona Supercom- where they run. It should also provide support for dis- puter Center). The “Related Work” sidebar describes some tributed file management, user authentication, resource of these systems and how they differ from Ibis. management, and interoperability between different Ibis has won prizes in numerous international com- middleware systems. petitions including the International Scalable Computing

56 COMPUTER

Challenge at CCGrid 2008 (for scalability), the International addresses. The SmartSockets library automatically solves Data Analysis Challenge for Finding Supernovae at IEEE such problems using existing and novel solutions, includ- Cluster/Grid 2008 (for speed and fault tolerance), and the ing reverse connection setup, overlay routing, and Secure Billion Triples Challenge at the 2008 International Seman- Shell (SSH) tunneling. tic Web Conference (for general innovation). SmartSockets creates an overlay network with a set of interconnected support processes called hubs. Em- IBIS PROGRAMMING SYSTEM ploying gossiping techniques, the hubs discover which The Ibis programming system provides many program- other hubs they can connect to, using SSH tunneling ming models, all implemented on top of the IPL. if necessary. This overlay network can be used to help solve connectivity problems or route the application’s IPL network traffic. The IPL is a Java-based communication library that When creating a connection, SmartSockets initially tries provides robust communication and resource-tracking to set up a regular (direct) TCP connection. If this fails, it mechanisms. It typically ships with the application as jar uses the overlay network to send a request for a reverse (Java archive) files, so no additional preinstalled libraries connection setup to the target machine. If this also fails, need be present at any destination machine. the library creates a virtual connection that uses over- The library provides a range of communication primitives including those for point-to-point and multicast communication. It supports streaming communication, which is especially important in high-latency environments The IPL has been designed as this allows overlapping of serialization, communication, specifically for real-world distributed and deserialization of data. The IPL avoids copying over- environments where resources can head as much as possible and uses bytecode rewriting to be added or removed dynamically. generate efficient serialization and deserialization functions. Consequently, it can significantly outperform Sun remote method invocation (RMI) communication and even C/MPI, in particular for complex data structures.4 lay routing. SmartSockets also handles machines with The IPL has been designed specifically for real-world multiple network addresses via multihoming. During distributed environments where resources can be added or connection setup, it considers all source and target ad- removed dynamically. It incorporates a mechanism, Join- dresses. It uses heuristics to determine the combinations Elect-Leave (JEL), that tracks which resources are being most likely to work and then tries these first. Extra identity used and what roles they have. JEL is based on the concept checks in the procotol ensure that it reaches the correct of signaling: it notifies the application or runtime system target machine. A worldwide experiment5 demonstrated when resources are added to or removed from the compu- the library’s effectiveness: in 30 realistic connectivity sce- tation. To select resources with a special role, it includes narios, SmartSockets was always capable of establishing elections. JEL thus provides the building blocks for fault a connection, while traditional sockets only worked in six tolerance and malleability by giving an up-to-date view of these. of available resources, allowing applications and runtime Ibis programming models systems to respond to changes when required. A number of pure-Java IPL implementations are avail- Besides the IPL, Ibis provides a range of programming able using the Ibis SmartSockets library, TCP (Transmission models, from low-level message passing to high-level di- Control Protocol), UDP (User Datagram Protocol), and vide-and-conquer parallelism. It implements the following Bluetooth. In addition, we provide implementations using programming models using the IPL: specialized non-Java libraries, such as MX (Myrinet) or MPI. The SmartSockets, TCP, and UDP implementations also t MPJ, the MPI binding for Java; work on the Android smartphone platform. t Satin, a divide-and-conquer model;6 t RMI, object-oriented remote procedure call; SmartSockets t group method invocation (GMI), a generalization of Running a parallel application on distributed resources RMI to group communication, including multicast and is complicated due to connectivity problems that make all-to-all communication; direct communication difficult or impossible. Incoming t Maestro, a fault-tolerant and self-optimizing dataflow traffic at a node may be restricted by a firewall or NAT. The model;7 and presence of multiple network interfaces and IP addresses t Jorus, a programming model for data-parallel multi- can cause addressing problems, as can private network media applications.8

AUGUST 2010 57

RESEARCH FEATURE

The higher-level programming models’ runtime sys- technique dynamically forwards application calls on the tems exploit the IPL and JEL to address many distributed JavaGAT API to one or more middleware adaptors that programming difficulties. A good example is Satin, which implement the requested functionality. Selection occurs uses JEL resource-tracking mechanisms to make applica- at runtime and uses policies and heuristics that auto- tions malleable and fault tolerant. For example, Satin can matically select the best available middleware, enhancing reexecute subtasks if a processor crashes. Also, it can dy- portability. If an operation fails, the intelligent-dispatch- namically schedule subtasks on new machines that become ing feature will automatically select and dispatch the API available during the computation, and it can migrate sub- call to an alternative middleware. This process continues tasks if machines leave the computation. Satin’s scheduler until a middleware successfully performs the requested also does locality optimizations: divide-and-conquer pro- operation. grams are inherently hierarchical and can therefore be Although such flexibility comes at the cost of some run- mapped efficiently onto a hierarchical wide-area system time overhead, this is often negligible compared to the such as a grid. Likewise, the scheduler does latency hiding; cost of the operations themselves. For instance, a Globus if it needs to retrieve new jobs from distant machines, it will job submission takes several seconds, while the overhead do that asynchronously, without blocking.6 introduced by the JavaGAT is less than 10 ms. However, the additional semantics of the high-level API can introduce some overhead. For instance, if a file is copied, the JavaGAT The JavaGAT uses intelligent first checks if the destination already exists or is a directory. These extra checks cost time because they require dispatching to integrate multiple remote operations. middleware systems with different The JavaGAT API isn’t the lowest common denominator and incomplete functionality into a of the underlying middleware APIs. Instead, the JavaGAT single, consistent system. offers rich default functionality and can combine features of multiple middleware layers with its intelligent-dispatching technique. Consider the following real code example, Satin thus makes these problems transparent to the user which copies remote files and entire directories between and application. Applications written with a lower-level sites: programming model like the IPL must deal with such problems explicitly, but Ibis gives them the necessary low-level 1 import org.gridlab.gat.*; mechanisms to do so. 2 import org.gridlab.gat.io.File; IBIS DEPLOYMENT SYSTEM 3 4 public class Copy { The Ibis deployment system consists of a software stack 5 public static void main(String [] args) and a graphical user interface for deploying and manag- throws Exception { ing applications. The GUI is implemented on top of the 6 GATContext context = new JavaGAT. GATContext(); JavaGAT 7 URI source = new URI(args[0]); Writing distributed applications using existing mid- 8 URI dest = new URI(args[1]); dleware APIs is a daunting task. APIs change frequently 9 and are often incomplete and too low-level.9 The JavaGAT 10 File file = GAT.createFile(context, provides a high-level API that facilitates development of source); // Create a GAT file complex applications. This API is object oriented and 11 offers high-level primitives for access to the distributed 12 file.copy(dest); // The actual file system, independent of the middleware that implements or directory copy this functionality. The primitives provide access to remote 13 data, start remote jobs, and support monitoring, steering, Shutdown the JavaGAT user authentication, resource management, and storing of 14 GAT.end(); // application-specific data. The JavaGAT uses an extensible 15 } architecture, wherein adaptors (plugins) provide access to 16 } the different types of middleware. The JavaGAT also uses intelligent dispatching to integrate This code is middleware independent and demonstrates multiple middleware systems with different and incom- the JavaGAT API’s power; equivalent Globus code would plete functionality into a single, consistent system. This take hundreds of lines. The JavaGAT allows programmers

58 COMPUTER

to ignore system-level peculiarities and instead focus on user can add resources to a running application by simply domain-specific problems. providing contact information such as a host address and The JavaGAT doesn’t provide a new user/key manage- user credentials. This information can be reused in later ment infrastructure. Rather, its security interface provides experiments. generic functionality to store and manage security in- APPLICATION EXAMPLE: formation such as usernames and passwords. Also, the MULTIMEDIA CONTENT ANALYSIS JavaGAT provides a mechanism to restrict the availability of security information to certain middleware systems or We illustrate Ibis with an application that performs real- remote machines. Currently, the JavaGAT supports many time recognition of everyday objects. Images produced by different middleware systems such as Globus, UNICORE a camera are processed by an advanced algorithm that (Uniform Interface to Computing Resources), gLite, PBS extracts feature vectors from the video data, which de- (Portable Batch System), SGE (Sun Grid Engine), KOALA scribe local properties like color and shape. To recognize (a co-allocating grid scheduler), SSH, GridSAM, Amazon an object, the application compares the object’s feature EC2 (Elastic Compute Cloud), ProActive, GridFTP, HTTP, vectors to ones stored earlier and annotated with a name. SMB (Server Message Block)/CIFS (Common Internet File System), and Zorilla. Zorilla IbisDeploy provides a simple and generic API and GUI that can Most existing middleware APIs lack coscheduling automatically perform commonly capabilities and don’t support fault tolerance and malle- used deployment scenarios. ability. To overcome these problems, Ibis provides Zorilla, a lightweight P2P middleware that runs on any real-world distributed system. In contrast to traditional middleware, Zorilla has no central components and is easy to set up As this is a compute-intensive problem with soft real- and maintain. It supports fault tolerance and malleability time constraints, a large distributed system performs the by implementing all functionality using P2P techniques. analysis.8 A data-parallel application running on a single If resources used by an application are removed or fail, site processes a single video frame. Calculations over con- Zorilla can automatically find replacement resources. It secutive frames are distributed over different sites in a was specifically designed to easily combine resources in task-parallel manner. multiple administrative domains. The initial application was written in C++/MPI, using To create a resource pool, a Zorilla daemon process TCP and SSH tunnels for wide-area communication. This must be started on each participating machine. Also, each program used manual deployment, was vulnerable to machine must receive the address of at least one other connectivity problems and partial failures (each caus- machine to set up a connection. Jobs can be submitted to ing the entire application to fail), and was frustrating Zorilla using the JavaGAT or, alternatively, using a com- to write and maintain on heterogeneous hardware and mand-line interface. Zorilla then allocates the requested middleware. Step by step, we replaced all its components number of resources and schedules the application, taking with an implementation in Java and Ibis. With the new user-defined requirements like memory size into account. program, the IbisDeploy GUI makes deployment easy, The combination of virtualization and P2P techniques thus SmartSockets automatically corrects the connectivity makes it very easy to deploy applications with Zorilla. problems, and the JavaGAT accommodates the middleware heterogeneity. We provided robustness by adding IbisDeploy fault tolerance and malleability support to the application Many applications use the same deployment process. using the IPL-provided mechanisms. Therefore, IbisDeploy provides a simple and generic API The resulting application is compiled on a desktop ma- and GUI that can automatically perform commonly used chine and easily deployed onto a distributed system. It deployment scenarios. For example, when a distributed concurrently uses up to 20 clusters, commonly employ- Ibis application is running, IbisDeploy starts the Smart- ing a total of 500 to 800 cores, and a mix of different Sockets hub network automatically. It also automatically middleware. The application even runs on the Android uploads the program codes, libraries, and input files (pre- smartphone platform, allowing distributed object recogni- staging) and automatically downloads the output files tion from mobile devices. (poststaging). EXPERIMENTAL EVALUATION The IbisDeploy GUI, shown in Figure 3, lets a user start, monitor, and stop applications in an intuitive manner and To evaluate Ibis’s functionality and performance, we run multiple distributed applications concurrently. The carried out a series of experiments with the multimedia

AUGUST 2010 59

RESEARCH FEATURE

Figure 3. The IbisDeploy GUI lets users load applications and resources (top middle) and keep track of running processes (bottom half). The top left of the figure shows the locations of available resources; the top right shows the SmartSockets network consisting of hubs and compute nodes. A video presentation is available at www.cs.vu.nl/ibis/demos.html.

application (see www.cs.vu.nl/ibis/demos.html for a video for task-parallel processing of subsequent images. demonstration). We used the Distributed ASCI Supercom- We used IbisDeploy to start a client on a local machine puter 3 (DAS-3), a five-cluster distributed grid system in and to deploy four data-parallel multimedia servers, each the Netherlands; additional clusters in Chicago, Japan (the on a different DAS-3 cluster (using 64 machines in total). Chiba and Tsukuba InTrigger sites), and Sydney; the US East All code was implemented in Java/Ibis, compiled on the Amazon EC2 cloud system; and a desktop grid and single client machine, and deployed directly from there. No ap- stand-alone machine, both in Amsterdam. Together, these plication software or libraries were initially installed on machines comprised a real-world distributed system. any other machine. We first compared the performance of Java/Ibis and Using a single multimedia server resulted in a process- C++/MPI implementations of the data-parallel processing rate of approximately 1.2 frames per second. The ing of a single video frame. On a single machine the Java simultaneous use of two and four clusters led to linear program is about 12 percent slower than the C++ version, speedups at the client side of 2.5 and 5 frames per second, which is within acceptable limits for a “compile once, run respectively. Adding additional clusters such as an EC2 everywhere” application executing inside a virtual ma- cloud, a local desktop grid, and a local stand-alone machine. On an 80-node DAS-3 cluster with the Myri-10G chine improved the frame rate even further. We thereby (Myricom 10-Gbit Ethernet) local network, the Java/Ibis and obtained a worldwide system using a variety of grid mid- C++/MPI programs have very similar speedup (scalability) dleware—Globus, Zorilla, and SSH—simultaneously from and communication overheads. within a single application. In the distributed version of our application, the data- The SmartSockets hub network circumvented a range parallel analysis is wrapped in a multimedia server. Client of connectivity problems between the sites. Many sites applications can upload an image or video frame to such a have a firewall, and the Japan and Australia clusters can server and receive back a recognition result. When multiple only be reached using SSH tunnels. In addition, almost all servers are available, a client can use these simultaneously of the applied resources have more than one IP address.

60 COMPUTER

SmartSockets automatically selected the appropriate ad- bis drastically reduces the effort needed to create dresses when two sites communicated. Without it, only and deploy applications for real-world distributed the DAS-3, desktop grid, and stand-alone machine would systems that consist of ad hoc combinations of have been reachable. clusters, grids, clouds, desktop grids, stand-alone To test Ibis’s fault-tolerance mechanisms, we conducted I machines, and even mobile devices. To achieve this, an experiment in which an entire multimedia server it integrates solutions to many fundamental distributed crashed. The resource-tracking system noticed this crash computing problems in a single modular programming and signaled the application. The client then removed the and deployment system, written entirely in Java. crashed server from the list of available servers. The ap- An important lesson learned from Ibis is that resource- plication continued to run, with the client forwarding video tracking functionality is as essential as communication frames to the remaining servers. functionality. While communication is among the basic We also accessed the multimedia servers from an HTC capabilities of any distributed programming system, Ibis T-Mobile G1 smartphone, which used Ibis to upload pic- is one of the few systems that support resource tracking tures taken with the phone’s camera and receive back a to implement fault tolerance and malleability. A second recognition result. Running the full application on the important lesson is that direct, two-way connectivity smartphone itself isn’t possible due to CPU and memory is rare in a real-world distributed system. However, limitations. Using the multimedia servers, however, the SmartSockets achieves this in a transparent manner. phone obtained a result in about three seconds. This Another lesson is that, for portability, it’s not advisable clearly shows Ibis’s potential to open up mobile comput- to implement applications using one particular middle- ing to compute-intensive applications. Using IbisDeploy, it’s ware system but to use a middleware-independent API, even possible to deploy the entire distributed application such as the JavaGAT, instead. Ibis also tries to make dis- from the smartphone itself. tributed programming easier by providing high-level programming models on top of these mechanisms. Satin, OPEN PROBLEMS AND FUTURE WORK for example, makes fault tolerance and malleability The Ibis programming subsystem is mostly useful for transparent and automatically performs locality and applications written in Java or languages that compile to latency-hiding optimizations. Java source code or bytecode. Java applications can use Ibis can be downloaded for free at www.cs.vu.nl/ibis. non-Java libraries through the Java Native Interface and invoke non-Java executables with the Process.exec method. References Theoretically, non-Java applications could also use the IPL 1. I. Foster, C. Kesselman, and S. Tuecke, “The Anatomy of through the JNI, but this is complicated. Despite these re- the Grid: Enabling Scalable Virtual Organizations,” Int’l J. strictions, many applications have been programmed on High-Performance Computing Applications, Aug. 2001, pp. top of Ibis. In addition, the Ibis deployment subsystem has 200-222. been used to deploy both Java and non-Java applications. 2. D. Butler, “The Petaflop Challenge,” Nature, 5 July 2007, We’re currently researching how to integrate support for ac- pp. 6-7. celerators like GPUs, which requires access to non-Java code. 3. K. Verstoep et al., “Experiences with Fine-Grained Distrib- In addition, existing high-level programming models uted Supercomputing on a 10G Testbed,” Proc. 2008 8th IEEE Int’l Symp. Cluster Computing and the Grid (CCGrid don’t cover all application domains, leaving some appli- 08), IEEE CS Press, 2008, pp. 376-383. cations to use the IPL directly because they must address 4. R.V. van Nieuwpoort et al., “Ibis: A Flexible and Efficient locality optimizations, fault tolerance, or malleability Java-Based Grid Programming Environment,” Concurrency themselves. We’re thus developing more flexible runtime and Computation: Practice and Experience, June 2005, pp. support in Ibis for a broader range of high-level program- 1079-1107. ming models. 5. J. Maassen and H.E. Bal, “SmartSockets: Solving the Con- nectivity Problems in Grid Computing,” Proc. 16th Int’l Finally, the interoperability layer (JavaGAT) introduces Symp. High Performance Distributed Computing (HPDC some runtime overhead. In practice, this overhead is in- 07), ACM Press, 2007, pp. 1-10. significant except for operations that provide additional 6. R.V. van Nieuwpoort et al., “Satin: A High-Level and Effi- semantics such as remote error checks. More importantly, cient Grid Programming Model,” ACM Trans. Programming the JavaGAT’s intelligent-dispatching technique leads to Languages and Systems, Mar. 2010, article no. 9. more complex error reporting and debugging if opera- 7. C. van Reeuwijk, “Maestro: A Self-Organizing Peer-to-Peer tions fail. Instead of a single error message, the user now Dataflow Framework Using Reinforcement Learning,” Proc. 18th ACM Int’l Symp. High Performance Distributed gets one error message per middleware layer that the Ja- Computing (HPDC 09), ACM Press, 2009, pp. 187-196. vaGAT attempted to use. Visual debugging and profiling 8. F.J. Seinstra et al., “High-Performance Distributed Video tools should be developed to help the user address these Content Analysis with Parallel-Horus,” IEEE MultiMedia, problems. Oct. 2007, pp. 64-75.

AUGUST 2010 61

RESEARCH FEATURE

9. R. Medeiros et al., “Faults in Grids: Why Are They So Bad Nick Palmer is a PhD student in the Department of Computer and What Can Be Done About It?” Proc. 4th Int’l Workshop Science at Vrije Universiteit. Contact him at [email protected].______Grid Computing (GRID 03), IEEE CS Press, 2003, pp. 18-24. Gosia Wrzesinska is a senior software engineer at VectorWise, 10. R.V. van Nieuwpoort, T. Kielmann, and H.E. Bal, “User- where she works on high-performance query processing for Friendly and Reliable Grid Computing Based on Imperfect database engines, and received a PhD in computer science Middleware,” Proc. 2007 ACM/IEEE Conf. Supercomputing from Vrije Universiteit. Contact her at [email protected].______(SC 07), ACM Press, 2007, article no. 34. Thilo Kielmann is an associate professor in the Department of Henri E. Bal is a full professor in the Department of Computer Computer Science at Vrije Universiteit and a steering group Science, where he heads the High Performance Distributed member of the Open Grid Forum and Gridforum Nederland. Systems research group, at Vrije Universiteit, Amsterdam, Contact him at [email protected].______the Netherlands. Contact him at [email protected]. Kees van Reeuwijk is a postdoctoral researcher in the Depart- Jason Maassen is a postdoctoral researcher in the Depart- ment of Computer Science at Vrije Universiteit. Contact him ment of Computer Science at Vrije Universiteit and one of at [email protected].______the original designers of the IPL and SmartSockets. Contact him at [email protected]. Frank J. Seinstra is an assistant professor in the Department of Computer Science at Vrije Universiteit. Contact him at Rob V. van Nieuwpoort is a postdoctoral researcher in the [email protected]. Department of Computer Science at Vrije Universiteit and ASTRON (Netherlands´ Institute for Radio Astronomy) and one Ceriel J.H. Jacobs is a scientific programmer in the Department of the original designers of the IPL and the JavaGAT. Contact of Computer Science at Vrije Universiteit and the maintainer him at [email protected].______of the Ibis software. Contact him at [email protected].______Niels Drost is a postdoctoral researcher in the Department Kees Verstoep is a scientific programmer in the Department of Computer Science at Vrije Universiteit and the original of Computer Science at Vrije Universiteit. Contact him at designer of Zorilla. Contact him at [email protected][email protected].______Roelof Kemp is a PhD student in the Department of Computer Science at Vrije Universiteit. Contact him at [email protected].______Timo van Kessel is a PhD student in the Department of Com- Selected CS articles and columns are available for free at puter Science at Vrije Universiteit. Contact him at [email protected].______http://ComputingNow.computer.org. “All writers are vain, selfi sh and lazy.” —George Orwell, “Why I Write” (1947) (except ours!)

The world-renowned IEEE Computer Society Press is currently seeking authors. The CS Press publishes, promotes, and distributes a wide variety of authoritative computer science and engineering texts. It offers authors the prestige of the IEEE Computer Society imprint, combined with the worldwide kYd]kYf\eYjc]laf_hgo]jg^gmjhYjlf]j$l`]k[a]flaÚ[Yf\ technical publisher Wiley & Sons. For more information contact Kate Guillemette, Product Development Editor, at [email protected].

______XXXDPNQVUFSPSHDTQSFTT

62 COMPUTER

RESEARCH FEATURE Simulating the Universe on an Intercontinental Grid

Simon Portegies Zwart, Leiden Observatory Cees de Laat, University of Amsterdam Tomoaki Ishiyama, University of Tokyo Stephen McMillan, Drexel University Derek Groen, Leiden Observatory Kei Hiraki, University of Tokyo Keigo Nitadori, RIKEN Stefan Harfst, Leiden Observatory Junichiro Makino, National Astronomical Paola Grosso, University of Amsterdam Observatory of Japan

The computational requirements of simulating a sector of the universe led an international team of researchers to try concurrent processing on two supercomputers half a world apart. Data traveled nearly 27,000 km in 0.277 second, crisscrossing two oceans to go from Amsterdam to Tokyo and back.

cientists’ understanding of the universe is These models have been quite effective and enabled much hampered by the elusive nature of its largest progress in the last decade. constituent, cold dark matter. The observa- Most simulations focus on the formation and evolution of large structures has provided some tion of the universe’s visible superstructure: galaxies, S insights into this phenomenon, particularly clusters, and dark matter. According to the / Cold Dark in understanding how dark matter behaves. Research- Matter model1 (where / stands for dark energy), the uni- ers know, for example, that the Milky Way and other verse is about 13.7 billion years old and comprises roughly galaxies rapidly rotate and that dark matter keeps the 4 percent baryonic matter (stars and gas), about 23 per- stars in these galaxies from being fiung out into space. cent nonbaryonic (cold dark) matter, and 73 percent dark They have learned that gravity affects dark matter, and energy.2 that these particles do not interact electromagnetically. The precise nature of dark matter remains a mystery. Although such observations continue, the information The most promising explanation is that dark matter is a they provide is limited. Thus, scientists rely on simulation collection of weakly interacting massive particles.3 Propo- to study dark matter in more depth and use the results nents of this theory view the entire universe as filled with from these runs to understand observations and make a finely grained substance that behaves like a gravitational predictions. There is (yet) no way to simulate the entire fluid but is otherwise invisible. The strength of the gravi- universe, so models capture only a sector using periodic tational force scales inversely with the distance squared, boundary conditions to mimic that sector ad infinitum. and because the enclosed mass scales with the cube of

RESEARCH FEATURE

the distance, the gravitational force becomes increasingly shows that performance would be satisfactory running dominant at larger scales. With no shielding mechanism, our cosmological N-body simulation on 10 supercom- all objects have gravitational interactions regardless of puters (up to 100 processors) in a ring topology with a their separation distance. high-bandwidth network. We are currently planning an- A popular technique for simulating dark matter is the other experiment using the CosmoGrid, this time with four gravitational N-body simulation, in which each simulated supercomputers. particle represents the fluid of dark matter particles. Such INITIAL CHALLENGES simulations can be resource-intensive, however, making it problematic to secure enough processing time to obtain Instead of starting the simulation at one location and meaningful results. A large-scale N-body simulation, for then switching halfway through the calculations to the example, can take several months on a single supercom- other machine, we ran the simulation on both super- puter, and that time is not easy to acquire. computers concurrently. At one point, the intricacies and politics of coupling the computers with an uncongested network and of resource scheduling made us wonder if the It is considerably easier to get less grid idea was worth the effort. Preparing for the concurrent time on many supercomputers, and simulation took about a year. current obstacles should be less On the other hand, running the simulation on a single supercomputer would have required its entire capacity formidable once high-performance for several months, which the institution probably would grid computing goes mainstream. not have granted. It is considerably easier to get less time on many supercomputers, which is a compelling reason to have a larger grid. Of course, the political and tech- COSMOGRID nical issues we faced initially would still be formidable, This administrative issue and the growing interest in re- but we believe that they will become less of an obstacle source distribution in large-scale scientific computing4 led once high-performance grid computing goes mainstream. us to explore the feasibility of using a computing grid for a Most technical complications stemmed from the required cosmological N-body simulation. The cost-effectiveness of hybrid parallelization strategies, diversity in topologies, distributing resources is even greater when computation scheduling, and load balancing. Different hardware ar- time scales more steeply than communication time,5 as it chitectures also complicated our experiment, particularly did in our case. since parts of the code were written in assembly language Our grid, which we named CosmoGrid, consisted of two and therefore machine dependent. supercomputers: an IBM Power6 at the National Academic Network topology Supercomputer Center of the Netherlands (SARA) in Am- sterdam and a Cray XT4 at the Center for Computational Figure 1 shows the network topology for the CosmoGrid Astrophysics of the National Astronomical Observatory of setup. Because neither supercomputer is directly con- Japan in Tokyo. Our goal in creating the grid was to use a nected to the optical network, communication occurs large number of processors efficiently even though they through a special node connected to the outside world were separated by half the planet, avoiding interproces- with a 10-GbE optical switch in Amsterdam and to a sor and intercontinental communication bottlenecks. If Neterion network interface controller in Tokyo. For each we could run the simulation successfully between two communication step, the processor identified the particles supercomputers separated by such a great distance, we to be communicated, packed them, and sent them to the could likely upscale our virtual organization to include supercomputer’s internal communication node, which more than two supercomputers. subsequently sent the particle package to the commu- We started our simulation in May 2009 and concluded nication node outside the firewall. This communication it nearly 14 months later. Although we faced many man- node transmitted the package to the partner computer. agement, political, and technical issues, we were able The straight-line distance is 9,400 km, but the data actually to rewrite the code of the well-known treePM (for parti- traveled approximately 27,000 km because the network cle-mesh) simulation algorithm6,7 to achieve 90 percent links crisscross the Atlantic Ocean, the US, and the Pacific efficiency, relative to a run on a single supercomputer. Ocean. With fiber optics, data travels at the speed of light, These results are exciting because they prove the effec- covering the distance in 0.138 second, for a round-trip time tiveness of using a grid for high-performance computing. of 0.277 second. On the basis of our first simulation results on Cos- We realized network communication by configuring moGrid, we constructed a performance model to see how two virtual large-area networks (vLANs) to create two paths our simulation code would scale on larger grids. Our model connecting both supercomputers. One vLAN was for pro-

64 COMPUTER

Mitaka, Tokyo Seattle CFCA Cray-XT4 Pacific Northwest supercomputer Gigapop StarLight Chicago JGN2Plus Northwestern Univ. Chicago

IEEAF Layer2 net NLR Layer1 vLAN 3780 vLAN 3781 vLAN 3788 vLAN 3789 WaveNet NetherLight

NAOJ T-Lex

SINET3 vLAN 3780 vLAN 3781 vLAN 455 vLAN 456 Layer2 net Tokyo

Vlan mapping JGN2Plus Los Angeles

10GE LAN-PHY JGN2Plus 10G STM-64/ SARA, Huygens supercomputer 10GE WAN-PHY NTT-com Otemachi, Tokyo JGN2Plus KDDI Otemachi Tokyo Amsterdam IRNC Circuit

Figure 1. Network topology for the CosmoGrid experimental setup. There are two main network connections between JGN2plus and StarLight. One goes via L-LEX and Gigapop; the other goes directly from StarLight to JGN2plus. Both paths have a network bandwidth of 10 Gbps, carried over 10 GE LAN-PHY (heavy red lines) and 10 GE WAN-PHY (heavy black lines) segments depending on the type of interfaces at the various switching points.

duction traffic and runtime synchronization; the other The Research and Education Networks (RENs) provided was for testing and collecting the simulation data at the the links for free, motivated by their vision to advertise petabyte storage array in Amsterdam. Because both paths and broaden their services to the scientific community. were dedicated to our experiments, we were able to avoid (In the past few years, many RENs have adopted a hybrid the packet loss, packet reordering, and latency changes model for their architecture, expanding it to routed Inter- typical of congestion. net Protocol services.) Users then had access to dedicated In the second path, we had to execute vLAN translight paths in which communication occurs at lower layers lation in the Japan Gigabit Network (JGN) because the of the open system interconnection reference model. We chosen vLAN numbers were unavailable along the entire settled for a flat Ethernet path but still had to make the path length. The choice of an available vLAN number network reservations three weeks before each simulation for the end-to-end communication should have been restart, and then it generally took a week for the light trivial; but because our multidomain light path lacked paths to become fully operational. automatic configuration tools, some human intervention Despite these delays, our overall experience using light was necessary. The time needed to solve problems often paths was quite positive. We could not have performed depended on the time to respond to e-mail and telephone our calculation without this kind of network setup, and we calls, which became problematic because our collabora- certainly enjoyed the advantage of having unrestricted and tion spanned several time zones. Debugging link failure exclusive use of the links. at Layer2 in a multidomain, multivendor infrastruc- Network demand ture was impractical and extremely time-consuming. A self-healing or automatic setup would help future experi- Because our simulation shared the computational ments enormously. domain, we had to synchronize the positions and velocities

AUGUST 2010 65

RESEARCH FEATURE

Table 1. Results of cosmological N-body simulation.

No. of processors No. of mesh cells in (Amsterdam + Wall-clock time for Time to compute No. of particles one dimension Tokyo) one step (seconds) forces (seconds) Speedup

16,777,216 128 30 + 30 31.2 23.58 1.51 134,217,728 256 30 + 30 116.8 105.7 1.81 134,217,728 256 60 + 60 71.9 58.8 1.64 134,217,728 256 120 + 120 53.0 36.8 1.39 1,073,741,824 256 500 + 250 60.0 40.0 — 8,589,934,592 256 500 + 250 380.0 310.0 —

of all particles on both machines throughout the simula- communication topology, established at startup, used mul- tion. This requirement introduced a huge demand on the tiple Transmission Control Protocol (TCP) connections for network. Fortunately, we had to communicate only a small paths between supercomputers. Running a separate thread boundary layer between the two halves of the universe for each stream enabled the concurrent use of multiple and the layer nearest the periodic boundary. The amount streams. For the smaller runs, we used 16 TCP streams; for of data communicated per step was larger runs, 64 streams. For intracluster communication, we used MPI and single TCP streams. S = 144N2/3 + 4N 3 + 4N/S byte, (1) comm p r SIMULATION ENVIRONMENT N where N is the number of particles, p is the resolution of Simulations use numerous algorithms, but one of the

the mesh in one dimension, and Sr is the sampling rate, more popular is the treePM method, so named because it which was 5,000 particles per sample for the produc- combines a Barnes-Hut treecode to simulate short-range tion simulation. We needed the first term of Equation 1 forces and a particle-mesh method to simulate long-range to communicate the tree structure; the second term, to interactions.6,10 This hybrid algorithm performs better than exchange particles in the border region; and the last term, either treecode or particle-mesh alone because it resolves to exchange the mesh. The first term depends slightly on both short- and long-range forces reasonably well and redshift, z, and on the opening angle in the treecode, e, but scales to a large number of processors through optimized the term is accurate for e = 0.5. Redshift is the standard domain decomposition.11 We adopted the treePM code measure of time in cosmology, conveyed in terms of z. To GreeM,11 which Tomoaki Ishiyama and colleagues rewrote provide perspective, the universe came into being at z →' to run efficiently on the hardware we selected. The re- and z = 0 represents redshift at the current time. written code integrates equations of motion in co-moving During the run, dense dark-matter clumps might not be coordinates using the leap-frog scheme with a shared but distributed evenly across the computational domain, so variable time step. Performance is good and accuracy suf- to use supercomputing resources efficiently, load balanc- ficient when the time step is at least an order of magnitude ing is required. We met this requirement by guaranteeing smaller than the crossing time in the densest halo.10 Intro- that the calculation time per step was the same on each ducing a Plummer softening to the force that is comparable computer, in effect generating a variable boundary layer to the local distance between particles prevented spurious between the two computers. Initially both computers re- relaxation effects in the high-density clumps. solved exactly half the universe, but in due time one of the Because most simulation time is spent calculating the two tended to deal with a larger volume. interparticle gravitational forces, we used single-precision To handle the communication within each super- X86-64 Streaming SIMD Extension (SSE) to optimize that computer, we used domain decomposition through the operation. Specifically, we used the inverse square-root message passing interface (MPI).8 To ensure efficient and SSE instruction because we found that it provides the stable data transfer, we developed the parallel socket li- best speedup in calculating Newtonian gravity. We fur- brary, MPWide, which made communication outside the ther improved performance by minimizing RAM access local MPI domains much easier.9 Communication consisted through the 16 XMM registers for the force operations and of data transfers between the local computation nodes and by operating on pairs of two 64-bit floating-point words the local communication node, and data transfer over the concurrently.12 light path between the supercomputers. With these optimizations, the Power6 with symmetric We included MPWide in all intercluster communication, multiprocessing was, per core, about 4.2 percent slower providing an interface similar to a regular MPI. A static than the Intel-based Cray XT4. This difference was due

66 COMPUTER

to our use of in-line assembly for the x86 SSE ver- 10 sion and intrinsic functions for the Power AltiVec Total time Calculation time architectures. Communication time 1 Data storage time SIMULATION RESULTS Table 1 presents our simulation results. We performed simulations in a computational box of 0.1 30 megaparsecs (Mpc) per side, which is roughly equivalent to 100 million light years, starting with z # 65, a softening of 300 pc, and a e of 0.3 for 0.01 Wall-clock time (hours) Wall-clock z > 10 and of 0.5 for z 10. The speedup in Table 1 is the wall-clock time of a single computer divided by the wall-clock time of 0.001 the grid as a whole (in this case, two supercomputers). In the table, speedup is for a grid in which each computer has the same number of processors, so 0.0001 10 1 0.1 0.01 we omit the last two entries in that column because Redshift (z) speedup is not properly defined. Because we were interested in kiloparsec to Figure 2. Wall-clock time as a function of redshift (z) for a simulation megaparsec structures, we generated the initial with 16,777,216 particles. Wall-clock time includes the network wait dark-matter distribution at z = 65. We also assumed stages. Throughout the simulation, about 25 percent of the total time was spent in communication. that the relation between the velocity and the potential in a Zel’dovich approximation is the same as that in linear theory.13 Using multiscale Gaussian random fields, larger runs, the average throughput increased to about 208 we then realized the density field, which we described in Mbits per second because particle package size increases terms of its power spectrum and generated using MPgrafic, with the number of particles (from Equation 1). a parallel (MPI) version of grafic that generates cosmologi- Figure 3 presents the results of our simulation with cal initial conditions.14 16,777,216 particles at z = 0. Each of the two panels shows We further adopted cosmological parameters that are the universe as it is seen by the two supercomputers, with consistent with the five-year Wilkinson Microwave Anisot- the black parts on the right of the left panel (left on the ropy Probe results.15 For clarity we opted for the nearest right image) indicating the part of the universe that resides round values, which yielded a matter (including dark) den- on the other machine. sity (: ) of 0.3 and a dark-energy density (: ) of 0.7. The m / PERFORMANCE MODEL slope for the scalar perturbation spectrum (nS) was 1.0,

and the amplitude of fluctuations (V8) was 0.8. With these From our measurements with the two-supercomputer parameters, the universe is about 13.7 billion years old, CosmoGrid, we constructed a performance model with

and the Hubble constant (H0) is approximately 70 km per two main components, the calculation (tcpu) and the data second per Mpc. transfer over the grid. We used this equation to address As Table 1 indicates, we ran several realizations at communication overhead: different mass and spatial resolutions to measure the per-

formance before we completed a simulation to z = 0. tcomm = ntlat + Scomm/b, (2) The simulation with 16,777,216 particles took about 10

hours to complete from z # 65 to z = 0. Figure 2 shows the where tcomm is communication time, tlat is network latency,

wall-clock time of this run decomposed in the time spent Scomm is defined by Equation 1, b is network throughput, in calculation, communication, and storing the data to the and v is a parameter we introduced to represent the file system. Throughout the simulation, about 25 percent number of data exchanges per step on the network con- of the total time was spent in communication. We encoun- necting the supercomputers. In the model, v corrects for tered a few problems between z # 16 and z # 15, probably code inefficiencies that resulted in multiple transmissions related to packet loss in the network transport protocol to the other supercomputer. suite, which resulted in an additional performance loss Figure 4 presents the results of the performance model of a few percent. that reflect the characteristics of the adopted comput- During communication, data-transfer speed averaged ers, network, and software environment for simulating 21.1 Mbits per second with peaks of about 7 Gbits per 8,589,934,592 particles. We modeled two topologies: ring, second, with 16 to 26 Mbytes transferred per step. For the which is ideal, and star, which is suboptimal because

AUGUST 2010 67

RESEARCH FEATURE

(a) (b)

Figure 3. Final snapshot of our simulation with 16,777,216 particles at the current time (z = 0). The snapshot shows how the (a) Cray XT4 in Tokyo and the (b) Power6 in Amsterdam view the universe. The solid region represents the memory share at the other site. The unequal division of the simulation box is due to the higher particle density in the simulation slab currently allocated to the Cray XT4.

Ring, 128 processors was approximately a factor of 8. Running on Star, 128 processors 100 supercomputers, the speedup increased Ring, 1,024 processors Star, 1,024 processors only to a factor of 10. Our experiments with the performance 10 model showed that speedup is limited by the

bandwidth, whereas latency, tlat = 0.138 s with n = 6, poses no limiting factor to our simulations. Thus, reducing total latency

Speedup will barely improve performance, but improving throughput by a factor of 10 would let us use an order of magnitude more processors per supercomputer while still acquiring acceptable speedup. 0.1

10 10 100 Number of supercomputers ith our cosmological cold dark-matter simulation, we Figure 4. Results of our performance model simulating 8,589,934,592 have realized the dream of particles. We varied the number of supercomputers, network topology as a a scalable grid for high-per- ring or star, and number of processors. Speedup is the wall-clock time of one W formance computing that supercomputer divided by the wall-clock time for the grid overall (number of Ian Foster and Carl Kesselman envisioned supercomputers and processors). We adopted an average bandwidth of 200 in 2001.5 Indeed, our frustration at the time Mbps with a latency (vt ) of 0.828 second for the external (grid) communica- lat spent preparing for the simulation soon tion and 10 Gbps for internal communication. The diagonal line indicates the gave way to enthusiasm about using the ideal scaling. computational grid as a high-performance resource. The overhead associated with communication goes through a single site, leading to communication latency and throughput is small relative congestion. With the ring topology running on 10 super- to the calculation cost: less than 10 percent for the larg- computers with at most 128 processors each, the speedup est simulation. Our CosmoGrid experiment revealed that

68 COMPUTER

latency forms no bottleneck, even if the computers are Light Cisco on National LambdaRail, TransLight, StarLight, separated by half the planet. NetherLight, T-LEX, and Pacific and Atlantic Wave. Simulating the formation of large-scale structures in the universe requires phenomenal computer power. Even with fully optimized numerical methods and approxima- References tions, our simulation took nearly 14 months. Although 1. A.H. Guth, “Inflationary Universe: A Possible Solution to the Horizon and Flatness Problems,” Physics Rev. D, vol. researchers are not likely to beat the speed of light, we are 23, no. 2, 1981, pp. 347-356. optimistic that bandwidth will eventually improve, making 2. D.N. Spergel et al., “Three-Year Wilkinson Microwave An- high-performance grid computing very attractive in the isotropy Probe (WMAP) Observations: Implications for foreseeable future. Cosmology,” ApJS, vol. 170, no. 2, 2007, pp. 377-408. Our experiment provides important insights for future 3. B.W. Lee and S. Weinberg, “Cosmological Lower Bound simulations of large-scale structure formation using a grid. on Heavy-Neutrino Masses,” Physical Rev. Letters, vol. 39, One is that a ring network topology appears to be the best 1977, pp. 165-168. strategy for performing cold dark-matter simulations using 4. G. Hoekstra et al., “Towards Distributed Petascale Comput- ing,” Petascale Computing: Algorithms and Applications, a treePM code on a grid. Another is that, although it is con- D.A. Bader, ed., Chapman & Hall/CRC Computational Sci- siderably easier to obtain 100 processors for a year from a ence Series, 2008. dozen supercomputer centers (as opposed to using a single 5. I. Foster, C. Kesselman, and S. Tuecke, “The Anatomy of supercomputer for the same simulation), this strategy would the Grid: Enabling Scalable Virtual Organizations,” Int’l J. require drastic changes to the centers’ coscheduling policy. High-Performance Computing Applications, vol. 15, no. 200, In some ways, our experiment serves as a proof of con- 2001, pp. 200-222. 6. G. Xu, “A New Parallel N-Body Gravity Solver: TPM,” As- cept. In principle, our setup would have reduced wall-clock trophysical Application J. Series (ApJS), vol. 98, 1995, pp. time by almost a factor of two, relative to a single super- 355-366. computer, but we spent more than a year preparing and 7. K. Yoshikawa and T. Fukushige, “PPPM and TreePM optimizing the code and acquiring and scheduling the Methods on GRAPE Systems for Cosmological N-Body resources—none of which proved to be trivial. On the other Simulations,” Pub. Astronomical Soc. Japan (PASJ), 2005, hand, had we not run the simulation on a grid, we might vol. 57, no. 6, 2005, pp. 849-860. have waited much longer to run it, since acquiring 1,024 8. W. Gropp et al., “A High-Performance, Portable Implemen- tation of the MPI Message Passing Interface Standard,” processors on a supercomputer for six months through a Parallel Computing, vol. 22, no. 6, 1996, pp. 789-828. normal proposal would have been problematic. We realize 9. D. Groen et al., “Light-Weight Communication Library that much work remains to resolve such practical matters for Distributed Computing,” Computational Science and as scheduling issues, network acquisition, and cooperation Discovery, June 2010 (to be published). among supercomputer centers—each of which is a major 10. R. Hockney and J. Eastwood, Computer Simulation Using bottleneck. However, none of these are insurmountable, Particles, Adan Hilger Ltd., 1988. and we plan to tackle those very problems on a larger scale 11. T. Ishiyama, T. Fukushige, and J. Makino, “GreeM: Mas- in our next CosmoGrid experiment. sively Parallel TreePM Code for Large Cosmological N-Body Simulations,” PASJ, vol. 61, no. 6, 2009, pp.1319-1330. 12. K. Nitadori, J. Makino, and P. Hut, “Performance Tuning of N-Body Codes on Modern Microprocessors: I. Direct Inte- Acknowledgments gration with a Hermite Scheme on x86_64 Architecture,” We are grateful to Hans Blom, Maxine Brown, Andreas New Astronomy, vol. 12, no. 3, pp. 169-181. Burkert, Halden Cohn, Jonathan Cole, Tom Defanti, Jan 13. M. Joyce and B. Marcos, “Quantification of Discreteness Effects in Cosmological N-Body Simulations: II. Evolu- Eveleth, Katsuyuki Hasabe, Douglas Heggie, Wouter Huis- tion Up to Shell Crossing,” Physics Rev. D, vol. 76, no. 10, man, Piet Hut, Mary Inaba, Akira Kato, Walter Lioen, Kees 2007, article 103505; http://arxiv.org/PS_cache/arxiv/ Neggers, Breanndan Ó Nuáillán, Hanno Pet, Steven Rieder, ______pdf/0704/0704.3697v2.pdf. Joaching Stadel, Huub Stoffers, Yoshino Takeshi, Thomas 14. S. Prunet et al., “Initial Conditions for Large Cosmological Tam, Jin Tanaka, Peter Tavenier, Mark van de Sanden, Simulations,” ApJS, vol. 178, no. 2, 2008, pp. 179-188. Ronald van der Pol Alan Verlo, and Seiichi Yamamoto. We 15. E. Komatsu et al., “Five-Year Wilkinson Microwave An- also thank Ben Moore for offering us a bottle of Pommery isotropy Probe (WMAP) Observations: Cosmological 1998 Cuvée Louise Brut for the first relevant scientific Interpretation,” 2008, ArXiv e-prints; http://arxiv.org/ results coming from this simulation. This work was sup- ______abs/0803.0547. ported by NWO (grants 643.200.503 and 639.073.803), the NCF (project SH-095-08), QosCosGrid (EU-FP6-IST- Simon Portegies Zwart is a professor in computational astro- FET Contract 033883), NSF IRNC, NAOJ, NOVA, LKBF, and physics at Leiden Observatory (Sterrewacht Leiden) at Leiden the JSPS. We thank the network facilities of SURFnet, University, The Netherlands. His principal scientific interests Masafumi Oe; IEEAF; WIDE; Northwest Gigapop and the are high-performance computational astrophysics and the Global Lambda Integrated Facility (GLIF) GOLE of Trans- ecology of dense stellar systems. Portegies Zwart received a

AUGUST 2010 69

RESEARCH FEATURE

PhD in astronomy from Utrecht University, The Netherlands. terascale eScience applications. De Laat received a PhD in com- Contact him at [email protected]. puter science from the University of Amsterdam. Contact him at [email protected]. Tomoaki Ishiyama is a graduate student in Junichiro Makino’s computational astrophysics research group at the University Stephen McMillan is a professor of physics at Drexel Univer- of Tokyo. His research interests include cosmology, partic- sity. His principal research interests are high-performance ularly high-performance cosmological N-body simulations. computation and the astrophysics of star clusters and galactic Ishiyama received a PhD in astronomy from the University of nuclei. McMillan received a PhD in astronomy from Harvard Tokyo. Contact him at [email protected] University. Contact him at [email protected]. Derek Groen is a graduate student in Portegies Zwart’s com- Kei Hiraki is a professor of computer science at the University putational astrophysics research group at Leiden Observatory. of Tokyo. His research interests are parallel and distributed His research interests include high-performance grid comput- systems, high-performance computing, and networking. Hiraki ing and distributed gravitational N-body simulations. Groen received a PhD in physics from the University of Tokyo. Con- received an MSc in grid computing from the University of tact him at [email protected]. Amsterdam. Contact him at [email protected]. Stefan Harfst is a postdoctoral fellow in Portegies Zwart’s Keigo Nitadori is a postdoctoral fellow at RIKEN in Japan. computational astrophysics research group at Leiden Ob- His research interests are N-body and hydrodynamical sim- servatory. His research interests are stellar dynamics and ulations, performance tuning, and multicore architectures. high-performance computing. Harfst received a PhD in astro- Nitadori received a PhD in astrophysics from the University physics from the University of Kiel, Germany. Contact him at of Tokyo, where he took part in the CosmoGrid experiment. [email protected]. Contact him at [email protected]. Paola Grosso is a senior researcher in the system and network Junichiro Makino is a professor of astrophysics at the Na- engineering research group at the University of Amsterdam. tional Astronomical Observatory of Japan. His research Her research interests are modeling, programming, and vir- interests are stellar dynamics, large-scale scientific simula- tualization of networks for eScience applications. Grosso tion, and high-performance computing. Makino received a received a PhD in physics from the University of Turin. Contact PhD in astronomy from the University of Tokyo. Contact him her at [email protected]. at [email protected]. Cees de Laat is an associate professor of system and network engineering at the University of Amsterdam. His research Selected CS articles and columns are available for free at interests are optical/switched Internet for data transport in http://ComputingNow.computer.org.

0,,,+LZPNU ;LZ[VM*VTW\[LYZJV]LYZ [OL[VVSZ[LJOUPX\LZHUKJVUJLW[Z\ZLK[V KLZPNUHUK[LZ[LSLJ[YVUPJWYVK\J[OHYK^HYL HUKZ\WWVY[P]LZVM[^HYL+ ;PZHSLHKLYPU HUHS`ZPZVMJ\YYLU[HUKULHYM\[\YLWYHJ[PJL

^^^JVTW\[LYVYNKLZPNU______

70 COMPUTER

www.computer.org/election Click and Vote ELECTION Online!

IEEE Computer Society Election

Nominees for Computer Society Officers and Board of Governors Positions in 2011

On the following pages are the position statements and biographies of the IEEE Computer Society’s candidates for president-elect, first and second vice presidents, and Board of Governors. Within each category, candidates are listed in alphabetical order. Election of officers to one-year terms and of Board members to three-year terms, each beginning 1 January 2011, will be by vote of the membership as specified in the bylaws. Ballots must be returned no later than 12:00 noon EDT on Monday, 4 October. Members in all regions can vote via the Web at www.computer.org/election or by fax to Survey & Ballot Systems at +1-952-974-5110. Return ballots by mail to the IEEE Computer Society, c/o Survey & Ballot Systems, PO Box 46430, Eden Prairie, MN 55344-9876, USA. For replacement ballots, call +1-202-371-0101. Results will be announced in the December issue of Computer. The opinions expressed in the statements are those of the individual candidate and do not necessarily reflect Computer Society positions or policies.

NOMINEES FOR PRESIDENT-ELECT

Roger U. Fujii John W. Walz Position statement. The IEEE Computer Position statement. As a lifetime IEEE Society is the world’s preeminent society for member and a volunteer contributor to the computer science professionals, represent- IEEE Computer Society for over 25 years, ing the academic community and industry I have been an advocate for membership practitioners. Society members represent a and increasing its value for our members as vast resource of the best talent and intellec- well as their employers. As president, I will tual property (IP) in the profession. In a time continue working to advance the theory and when information access and exchange are application of computer and information- moving at Internet speed, the Society needs to adapt its communi- processing technology through growth of member engagement cation and information access channels (communities of interest) and affiliation and through access to timely, relevant, high quality, to give members access to IP easily and quickly. and affordable intellectual property in tailored formats. If elected president, I will work to reduce the cost of IEEE The Society should find ways to assist its members through Computer Society membership, maintain a viable Society finan- active engagement, helping all our members create, collaborate, cial structure, reinvigorate our new member drive, and develop critique, recommend, share, and preserve their intellectual prop- more agile channels (for example, instant community forums) erty. This includes our underserved markets of computing and for members to access the Society’s vast intellectual information. information-technology practitioners and the enormous mem- Specifically, my goals are to: bership talent and potential in Regions 9 and 10. As president, I will continue to explore ways to recruit and engage volunteers; let 1. increase CS membership by reaching out to university members define their organization; promote collaboration across students, practitioners in our corporate world, and new pro- conferences, publications, and standards; and implement new fessionals in Asia (Japan, China), the Middle East, Europe, content-generation and distribution. Eastern Europe, India, and South America; My career, volunteer experience, service on the Computer Soci- 2. develop new distribution channels to get Computer Society ety Executive Committee (formerly as VP for Standards and now services and products (such as ReadyNotes, book series, and as VP for Technical and Conference Activities) under four presi- seminars) to our members rapidly and at low cost; dents—Mike Williams, Rangachar Kasturi, Kathy Land, and Jim 3. provide new services and products to meet social engineer- Isaak—and vision for the IEEE Computer Society uniquely qualify ing challenges in smart grid, social networking, and green IT me for this leadership role. It is a leadership that coaches commit- issues; tees to improve their vitality, recognizes volunteer contributions, 4. manage with strong fiscal discipline and support workflow- makes plans that align with Society goals, collaborates across efficiency initiatives to provide agile and quality service to Society boards and IEEE units, invests in emerging technolo- our members; and gies, engages the practitioner community, listens to our members 5. encourage new initiatives focused on the emerging needs and key stakeholders, supports members’ career growth, and

(Fujii continued on next page) (Walz continued on next page)

(Fujii continued from page 71) (Walz continued from page 71)

and interests of Society members such as gaming technology, improves the member’s experience. It is a leadership that promotes cloud computing, and learning technology. decisions based on facts, encourages consensus, and keeps our activities aligned with our mission. It is a leadership based on I realize that these goals can only be achieved by the teamwork experience, skill, and accomplishment that continues the recent of Society volunteers, staff, the Board of Governors, and elected fiscal transformation and establishes a solid financial base for leaders. I know that achieving them is possible because of our past future investments to reach our potential markets and improve achievements. As USA chair of international software engineering our Society. standards group ISO/IEC JTC/SC7, I reinvigorated participation I ask for your continued engagement in your Society and tenfold in international standards. As Press Operations Committee request your support for my vision, which is elaborated upon at chair, I led the formulation of a new book distribution channel and www.johnwalz.com. increased Society revenues. I am proud of the Society and its members. I know that the Soci- Biography. John W. Walz retired from Lucent/AT&T with more ety will continue its preeminent stature as the society of choice for than 25 years of software and systems engineering-management the world’s computer science professionals. For more information, leadership experience. Walz is serving as the 2009–2010 IEEE visit http://www.fujii.us.com. Computer Society vice president of the Technical and Conferences Activities Board. He has served on the Society’s Board of Governors Biography. Roger U. Fujii is vice president of Network Commu- in positions as first vice president and vice president for stan- nication Systems at Northrop Grumman, where he is responsible dards activities. He remains a member of the Computer Society for three business units with annual sales of US$1.05 billion, con- Standards Activities Board, Software and Systems Engineering trols $42 million in annual R&D, and manages 2,300 employees. Standards Executive Committee, Electronic Products and Services He has financial, technical, and managerial responsibilities for Committee, Planning Committee, and Executive Committee. major communications and network planning; intelligence, sur- Walz has held leadership positions in national and interna- veillance, reconnaissance, and satellite systems. He is a software tional professional organizations, including the US Technical safety pioneer and was responsible for the software certification Advisory Group for the ISO committee working on ISO 9001, of more than 30 US Air Force and Navy nuclear weapons systems. American Society for Quality Electronics and Communications, Fujii received a BS in engineering and an MS in electrical engi- Quality Excellence for Suppliers of Telecommunications Forum, neering and computer science from the University of California, and Information Integrity Coalition. Berkeley. He also attended the UCLA John Anderson Management, Walz received an MS in electrical engineering from Ohio State Darden Management, and Harvard Business Management schools. University. He has coauthored three books on the use of IEEE Fujii’s IEEE and Computer Society activities include first vice Computer Society software engineering standards to support president and vice president of the Computer Society Standards CMMI, ISO 9001, and Lean Six Sigma. He is also a contributor to Activities Board (2010); Computer Society Board of Governors the IEEE Computer Society ReadyNotes and webinar programs. member (2006-2009); Computer Society chair for Std 1012–Soft- Walz participates in the Distinguished Visitor Program and is a ware Verification and Validation (1985-present); and Computer recipient of the Society’s Golden Core, Distinguished Service, and Society Press Operations Committee member (2003-2005). Meritorious Service Awards. He is an IEEE Fellow, a Computer Society Golden Core member, and a recipient of Computer Society Meritorious and Outstanding Service Awards. Fujii was named the Chinese Institute of Engineers Asian American Executive of the Year in 2009.

72 COMPUTER

NOMINEES FOR FIRST VICE PRESIDENT

Frank E. Ferrante David Alan Grier Position statement. Now serving a second Position statement. Service. Office in the term as an elected member of the Board of IEEE Computer Society is an opportunity to Governors, an active member of the Digi- be of service to the world, and, ultimately, tal Library Operations Committee, Society to the future. treasurer, and chair of the Electronic Prod- We serve the field by providing a venue ucts and Services Committee, I feel that if I for professional practice, a place where we were elected to serve as your Society’s vice can share best practices, validate knowl- president I could focus my experience on edge, and train the rising generation of developing improved cost-beneficial solutions to the problems leaders. We serve the world by reaching beyond our borders and we face in meeting today’s global educational needs. My interest disseminating our accomplishments to the organizations that can in being a volunteer has always been directed at helping mem- utilize them. We serve the future by sustaining an organization bers obtain the maximum benefit at a reasonable cost from our that can identify and follow new trends in our field. offerings. Our Society is preparing itself for change as it coopera- I have three approaches to these issues. First, I am a skilled tively integrates its operational services into that of our parent manager who works to make an organization operate well. operations of the IEEE. This effort is aimed at becoming more Second, I make efforts to communicate the accomplishments of technology agile and financially sound. I am dedicated to contin- our Society to our members and to the community that needs our ued change in the directions underway, while continuing to offer leadership. Finally, I work with our leaders to anticipate the needs the best to our members through our products and services. I feel for the coming years to prepare our society for its next steps. fortunate to have gained insight into the needs of our membership The three accomplishments that best illustrate this approach through my service to date, and I offer my pledge to support new are my efforts to promote electronic publications, my column in services and benefits to all of our members at competitive prices. I Computer, and work that I have been doing with other members to will continue to support positive advancements in our offerings as develop new organizational structures within the Society. These technology changes and always will be cognizant of our Society’s accomplishments point to my skills as a leader and my willing- cost of membership and product availability for training needs as ness to be of service to the Society, to the world, and to the future. appropriate for today’s members. A fuller statement of my positions, as well as podcasts that further explain my vision, can be found at http://sites.google.com/ Biography. Frank E. Ferrante, a life senior member of the IEEE ______site/dagrierieee. and member of both the IEEE Computer and Communications Societies, was elected to a second consecutive term on the Com- Biography. David Alan Grier is currently the vice president of puter Society Board of Governors for 2010-2012. He was awarded publications for the Computer Society and has spent much of the membership into the IEEE Computer Society’s Golden Core in 2009 past decade helping the Society develop new electronic products, and was appointed treasurer of the Computer Society for 2010. editing its periodicals, and writing for its members. He has served He also serves as chair of the Electronic Products and Services as editor in chief of IEEE Annals of the History of Computing, as Committee and is an active member of the Board of Governors’ chair of the Magazine Operations Committee, and as an editorial Planning Committee. Ferrante served as editor in chief and advi- board member of Computer. Grier currently writes the monthly sory board member of IT Professional and has served as a member column “The Known World” (www.computer.org/theknownworld). of various Society boards and committees, including Publications Outside the Society, he works as an associate professor of sci- and PAB-IT. He also served as chair of the IEEE-USA’s Medical ence and technology policy at George Washington University in Technology Policy Committee. Washington, DC, with a particular interest in policy regarding digital technology and professional societies (www.gwu.edu/~cistp). There, Grier has worked as a university administrator for the past 20 years and has demonstrated a capacity for organizational management. He served as leader of the undergraduate computer systems degree, director of the University Honors Program, assistant dean of engineering, and associate dean of International Affairs. Finally, Grier has worked extensively within the computer industry. He started as a programmer and systems designer for the old Burroughs Corporation. He has also worked extensively as a consultant in the field. A more detailed biography can be found at http://sites.google.com/site/dagrierieee/home/bio.

AUGUST 2010 73

NOMINEES FOR SECOND VICE PRESIDENT

Jon G. Rokne Sattupathu V. Sankaran Position statement. The Computer Soci- Position statement. The IEEE Computer ety is facing opportunities and challenges Society has just recorded its first year-to- in several areas. year increase in membership since the year 2000. This is a very positive sign, coupled t Membership. Developing new member- with the formation of the new MGA board, ship benefits is a high priority in view of with intent to align with IEEE’s MGA. This declining membership. This might is a crucial juncture for leading Computer include further incentives to attract new student members. Society activities and serving the Computer Retaining student members as full members should be a Society, its chapters, and members at large. This is also a time priority. when ACM is launching initiatives in countries like China, India, t Accreditation. I will work toward acceptance and recognition and Russia. The IEEE Computer Society therefore has to connect, of these accreditation efforts and enlist the help of IEEE-USA better than ever before with its members and potential members, in gaining government support for accreditation within the and demonstrate its true value and benefits to them. This can USA. happen by better engagement with the members and chapter t Conferences. Conferences provide the main forum for in-per- leaders; higher quality publications, technical activities, and con- son member contacts and in-person exchange of technical ferences; easier and more economical access to the Computer information. They are vital Computer Society activities. Society Digital Library; and better standards and professional Maintaining the viability of the conference program is there- services. It will be my pleasure to serve the Computer Society now fore a priority. and in coming years. t Internationalization. I consider it a priority to engage the international membership of the Computer Society more Biography. Sattupathu V. Sankaran received a BS in electrical effectively in Society activities. One way of doing this would engineering from Jadavpur University, India, and an MS in con- be to create online communities to discuss issues of interest trol systems from the University of Pennsylvania. He has worked to Computer Society members. in industry for more than 30 years, including IBM and Bharat t Publications. One of the main incentives for joining the Com- Heavy Electricals Limited (in India), Westinghouse, Electric Power puter Society has been its publishing program. Members Research Institute, and Duke Power (in the USA). Sankaran’s inter- typically can subscribe to IEEE and Society publications at ests focus on industry research and development, power plant a significantly reduced cost. The CSDL and IEL electronic controls, modeling and simulation, industry-academic relations, libraries have made individual subscription incentives of less and general management functions. He was a senior profes- value to many members due to institutional subscriptions. sor and associate dean at International Institute of Information I would support development of new publishing initiatives Technology-Bangalore before moving into IT consulting for Daim- that lead to member retention. lerChrysler, Infineon, Yahoo, and currently SAP Labs. t Open access has clear benefits for Computer Society mem- Sankaran received the Society for Computer Simulation’s bers and the community in general since it results in more Industry Technology Award in 1992 for his work on the EPRI freely available information. The challenge is to make open Mobile Training Simulator. access economically viable. A senior member of the IEEE, Sankaran served as IEEE Banga- lore Section chair in 2002-2003. Sankaran served as membership Biography. Jon G. Rokne is a member of the IEEE Computer development chair for Region 10 in 2004-2006 and helped to Society Board of Governors, a member of the Publications Board, revive the dormant Computer Society Bangalore chapter in 2008. and a member of the Society’s Audit Committee. He is also the He served as vice president for chapters in 2009 and is the vice vice president of the IEEE Publication Services and Products Board president of the new Member and Geographic Activities Board. (PSPB) and a member of the IEEE Board of Directors. He holds the IEEE Millennium Medal and became a Golden Core Rokne has completed two terms as vice president of publica- Member of the Computer Society in 2010. tions for the IEEE Computer Society and has served as a member of PSPB, PSPB Financial Committee, and PSPB Operations Commit- tee, also chairing a PSPB subcommittee on publications conduct. A Computer Society Golden Core member, Rokne has served as a member of the Publications Board, chair of the Transactions Operations Committee, and chair of an ad hoc committee for ReadyNotes. Rokne is a professor and former chair of the computer science department at the University of Calgary. He has published extensively in mathematics, including three jointly authored books. His main interests are interval analysis, global optimization, and computer graphics. Rokne has published in the areas of physically and biologically based computer simulations on leaves, auroras, ball lightning, and one jointly authored book, Light Interaction with Plants. In 2003, he organized the Pacific Graphics conference. For further information, see www3.telus.net/public/jrokne/ biography.html.______

74 COMPUTER

BOARD OF GOVERNORS NOMINEES (11 NOMINEES; VOTE FOR SEVEN)

Pierre Bourque Biography. Pierre Bourque is an associate professor and the Position statement. The IEEE Computer director of a professional master’s degree program in software Society must better address the needs of engineering at the École de technologie supérieure (ÉTS) of computing professionals in industry around the Université du Québec, Canada. He is a member of the IEEE the world. By virtue of my background in Computer Society’s Board of Governors (2010) and lead coedi- both industry and academia, I will advocate tor of the upcoming Guide to the Software Engineering Body of the vision that the Computer Society should Knowledge (SWEBOK Guide) v3. He is also coeditor of the 2001 and move from being a technical society to a 2004 versions of the SWEBOK Guide, sponsored by the IEEE Com- broader professional one. It needs to build puter Society and funded by numerous industrial partners. The upon its credibility and brand-name recognition to offer more SWEBOK Guide is also recognized as an ISO/IEC technical report. high-quality products and services that are current, practical, Bourque is currently a member of the Computer Society’s and relevant to professionals in the field, while maintaining its Professional Activities Board and its Software Engineering Com- long-standing tradition of technical excellence in the academic mittee. He served on the Professional Practices Committee from and research communities. Such products notably include bodies 2006 to 2009. He was a member of the Distinguished Visitor of knowledge, certification programs for professionals that are Program (2006–2009) and was the recipient of an Outstanding recognized by industry, as well as training delivered in various Contribution Award from the Computer Society in 2001. online formats and durations. This cohesive suite of products and Bourque received a PhD from the University of Ulster (Northern services must be actively promoted in the world’s marketplaces Ireland) on the topic of the maturation of the software engineering and meet the needs of professionals everywhere because comput- discipline and profession. Prior to his academic appointment, he ing is, without doubt, a global industry. was involved in software engineering, data modeling, and database design at the National Bank of Canada from 1987 to 1995.

José I. Castillo-Velázquez Strategic Planning Committee (2008-2011), technical administra- Position statement. The IEEE Computer tor of the Region 9 Virtual Regional Meeting (2010), and chair of Society needs to be more involved where the Region 9 Virtual Communities Committee (2008-2009). He jobs are produced, so we need to increase has worked for 15 years in the computer, electronic, and telecom- practitioners’ participation. It’s time to share munication industries as a practitioner in the public and private specific programs for collaboration among sectors (TELMEX 2006-2008, IFE-97 and DICI-94) as well as in all the regions. It’s time for pushing inter- private and public universities as a tenured professor at Universi- national programs that encourage baby dad Tecnologica de la Mixteca (1998-1999); at Universidad Popular boomers to share their experiences with Autónoma del Estado de Puebla (1999-2005), where he became young computer professionals in all countries. The Computer department chair; and at Benemérita Universidad Autónoma de Society should focus not only on new products and services, but Puebla (2005-2006). on the maximum potential use of the products and services that Since 2008, Castillo-Velázquez has been a tenured professor at it offers to its members today. It´s important to remember, the best Universidad Autónoma de la Ciudad de Mexico. He has authored technology is not always the newest one but the most adequate. 25 journal and conference papers. He has organized multidisci- Now, we need to review our Web platforms and their functional- plinary congresses and seminars and has participated in more ities in order to increase virtual collaboration to help everyone than 100 interviews for magazines, newsletters, radio, and televi- reduce cost and time. The Society must also pay even more atten- sion. A senior member of the IEEE since 2010, he is an independent tion to its impact on the environment. consultant for local companies and governments. Castillo- Velázquez received a BS in electronic sciences, with honors, and Biography. José Castillo-Velázquez is the editor in chief of the an MS in semiconductor devices from the University of Puebla. IEEE Region 9 newsletter (2008-2011), a member of the Region 9 Visit www.paginasprodigy.com.mx/a57852133 to learn more.

Dennis J. Frailey Biography. Dennis J. Frailey recently retired as a principal fellow Position statement. I have two goals as after a 35-year career at Raytheon and Texas Instruments. Along a BOG member — to increase the Society’s the way, he was a technical director, cycle-time expert, process support of students and working profes- improvement leader, head of many technical design teams, and sional members, and to enhance synergy speechwriter for company executives. Frailey has also maintained among various boards and committees. a distinguished career as an educator. At Raytheon, he created We need to reduce duplication of effort and and was master instructor for a company-wide software project better integrate and synchronize member management training program. Since 1978, he’s worked as an services and resources. As vice chair of the adjunct professor of computer science and software engineering Educational Activities Board, I led the development of our recently at the University of Texas and at Southern Methodist University, released online self-assessment courses in software engineering, where he cofounded the graduate software engineering program. which tie directly to our certification programs, the SWEBOK, and After earning a PhD in computer science from Purdue in 1971, related publications and standards. We’re working on additional Frailey joined SMU as assistant professor and rose to associate resources such as essential sets—collections of topic-focused with tenure before moving into computer architecture research at materials for individuals wishing to enhance their knowledge and Texas Instruments. Professionally, he was a regional representa- skills. I’d like to expand these efforts to include information tech- tive, council member, and vice president of ACM, and currently nology and computer engineering as well as important specialized serves in its Distinguished Lecturer program. Frailey is equally topics such as multicore processors and security. As a Board of active in the Computer Society, currently serving as a member of Governors member I’d be better able to focus the resources of the the Professional Activities Board and vice-chair of the Educational entire Computer Society on these efforts. Activities Board. An ABET program evaluator since 1987, Frailey recently won the ACM SIGCSE’s Lifetime Achievement Award in Computer Science Education.

AUGUST 2010 75

Atsuhiro Goto Biography. Atsuhiro Goto is vice president and general manager Position statement. I have no doubt that of the Cyber Space Laboratories at Nippon Telegraph and Tele- the IEEE Computer Society is the world’s phone (NTT). He is responsible for next-generation communication leading organization for computer profes- services that use a variety of innovative information technologies, sionals and, with its vast human network including the world’s top-level media processing technology and and commitment to professional excellence, open source software for future cloud computing. the most influential professional society in Goto has been with NTT R&D for more than 25 years and has Japan and Asian countries. made significant contributions on several cutting-edge business On the other hand, I am aware that the developments. His experience includes leadership of various Computer Society is now facing many challenges, especially in nationwide projects with major information and communication contributions to and from recent and emerging IT industries. technology vendors, and he currently manages Japan’s national The IT industry has been continuously transforming itself and project on “Highly Reliable Foundation Technology for Control of will expect the CS to contribute to new technologies, such as Cloud Computing Services.” Goto has also contributed to global cloud computing, which might change even the definition of standardization as vice chair of the Global Inter-Cloud Technol- “professionals.” ogy Forum. His research interests include cloud computing, IP On the basis of my 25+ years of industrial career experience networking, and new application architectures. and contributions to both IEEE and Japanese societies, I promise to Goto received a PhD from University of Tokyo in 1984. He is a make my best effort to bridge these challenges between the Com- member of the IEEE Computer Society; the ACM; the Institute of puter Society and the IT industry, including those in the emerging Electronics, Information, and Communications Engineers; and the industry communities in Japan and other Asian countries. Information Processing Society of Japan. He is an IPSJ fellow and served as a member of the IPSJ board from 2008 to 2009. He has been a member of the IEEE Computer Society Industry Advisory Board since 2009.

André Ivanov Biography. André Ivanov is head of the department of electrical Position statement. The Computer Soci- and computer engineering at the University of British Columbia. ety has been operating for years with a set He spent 1995-96 at PMC-Sierra and has held invited positions at of well-understood revenue sources—its the University of Montpellier II, the University of Bordeaux I, and conference and publications businesses. Edith Cowan University, in Perth, Australia. His primary research These enterprises are rapidly evolving. interests are in the area of VLSI, systems on a chip, and networks Rather than observing these evolutions on a chip. He has published more than 200 papers and holds four with a focus on reducing the Society’s cost US patents. Ivanov served as program chair of the 2002 VLSI Test of operations in response to the decreased Symposium and general chair of VTS 03 and VTS 04. In 2004, he revenues from these channels, the Computer Society should strive founded and co-chaired the first IEEE International GHz/Gbps Test to enable and facilitate new opportunities for its members. The Workshop and co-chaired the same event in 2005. In 2001, Ivanov big challenge remains in finding the products and services that cofounded Vector 12, a semiconductor IP company. members truly value, not only from North America but globally. Ivanov serves as associate editor for IEEE Transactions on We need to improve communication with members and engage Computer-Aided Design of Integrated Circuits and Systems, IEEE with volunteers so that they can more clearly see and leverage the Design & Test of Computers, and Kluwer’s Journal of Electronic benefits that the Society brings to their efforts and aspirations. If Testing: Theory and Applications. He served four years as chair of elected, I will work at making the Society shift its perspectives the IEEE Test Technology Technical Council. In 2008 he chaired to be even more global than it is today, and focused on offering the Computer Society Fellows Committee. Ivanov sits on the board increased value to its members. of directors of the IEEE Technology Management Council. He is a fellow of the Canadian Academy of Engineering and a registered professional engineer in British Columbia.

Paul K. Joannou Power Generation (formerly Ontario Hydro). He is a registered Position statement. There is great professional engineer and a longtime member of the IEEE. He opportunity to refocus the IEEE Computer has worked as a design engineer, manager of standards for all Society’s products and services to meet the engineering activities at Ontario Hydro’s nuclear plants, and an needs of computing professionals who have IT enterprise architect. He was responsible for developing the historically not been members and to better Canadian nuclear industry’s software engineering standards for satisfy the needs of existing members. This safety-related software. will require collaboration among all pro- Joannou has participated as a member and chair of software fessional organizations, both within the engineering standards working groups for the Canadian Standards Computer Society and IEEE and outside the IEEE, to deliver an Association, the ISO/IEC JTC1/SC7 committee responsible for soft- integrated set of products and services focused on member needs. ware and systems engineering standards, the IEC, the IEEE, and In my career, I have been fortunate to have gained experience the International Federation of Automatic Control. He has been a designing solutions at the software, systems, and enterprise levels. member of Canada’s Natural Sciences and Engineering Research Achieving strategic goals through alignment of cross-functional Council grant-selection committee, a member of Canadian Engi- organizations has been the focus of the latter half of my career. neering Accreditation Board accreditation teams, and a member As a member of the Board of Governors, I believe that I can make of the McMaster University Industrial Advisory Committee on a significant contribution to help the Computer Society achieve Software Engineering. He is currently chair of the IEEE Computer its strategic goals through effective alignment of all stakeholders. Society Professional Activities Board IT Committee. Joannou graduated from the University of Toronto with a BASc Biography. Paul K. Joannou has more than 30 years experi- in electrical engineering and an MEng in computer engineering. ence in software, system, and enterprise engineering at Ontario He is currently enrolled at McMaster University working toward a PhD in software engineering.

76 COMPUTER

Dejan S. Milojicic Read more at www.hpl.hp.com/personal/Dejan_Milojicic/ Position statement. If elected, I will: BoGNomination.______

1. better align the Computer Society with Biography. Dejan S. Milojicic is the founder and editor of Com- technology evolution by introducing puting Now. He has spent twenty years serving the Society in social networks, cloud computing, and positions that contribute to its vitality and have helped him to mobile access to the Computer Society. understand the Society’s needs and challenges. Milojicic is an IEEE As a part of Computing Now, I am already Fellow, a founding member of IEEE Distributed Systems Online, leading efforts that enable content deliv- and subsequently its editor in chief. He has been a member of the ery through FaceBook, LinkedIn, iPhone, Kindle, etc. IEEE Concurrency and IEEE Internet Computing editorial boards, 2. continue to strengthen membership through improved chair of the Technical Committee on Operating Systems, and educational, professional, and personal services. I will drive program chair of the IEEE Symposium on Agents, Systems, and community-based development of services and emphasize Applications. Milojicic has served on many program committees collaboration across sister societies. Given scarce IT resources, including IEEE ICAC, ICWS, ICDCS, and ICDE. services development is not easy and as a result, members In his career, Milojicic has led research that engaged him in cannot fully utilize Computer Society intellectual property. the global reach of the Computer Society and the needs of its 3. help evolve the Computer Society Special Interest Group. Com- members. At HP, he is responsible for projects from data center puter Society members can benefit from a dynamic technical management to pervasive computing. Currently, Milojicic is man- engagement that SIGs can offer. The ability to form a focused aging director of the Open Cirrus cloud computing testbed. He team of experts with selective access to CS digital library, has published two books, more than 50 papers, and holds eight collaboration tools, and newsletters will increase the value CS patents. Milojicic looks forward to bringing his extensive volunteer can offer to its members. and professional experience to the Board of Governors.

Paolo Montuschi a tenured full professor at Politecnico di Torino, Italy, where he Position statement. I feel very fortunate currently serves as an elected member of the Board of Governors, for the opportunity to put my professional chair of the Computer Engineering Department, and chair of the and management expertise at the service Board for Financial External Affairs. of the Computer Society. The experience For 20 years, Montuschi has been a member of IEEE and the gained over many years of volunteer activi- IEEE Computer Society, where he has served as a member-at-large ties has given me a unique insight into of the Computer Society’s Publications Board and as member of an the Society’s inner workings. I now truly IEEE ad-hoc Committee for Quality of Conference Articles in IEEE believe that, if elected, I can help the Board Xplore and of the Conference Publications Operating Commit- of Governors continue focusing on the opportunities offered by tee and Digital Library Operating Committee. He served as guest technology advancements to further improve the services to its editor and associate editor of IEEE Transactions on Computers members. from 2000–2004 and 2009–present, as well as program co-chair Introducing e-book publications, simplifying procedures and program committee member of several IEEE conferences. including conference sponsorship, and providing for member Montuschi authored many technical contributions published needs, benefits, and services will be among my top priorities, in peer-reviewed conferences, journals, IEEE transactions, and while addressing the current budget issues and working to main- Computer Society publications. He has been active in several sci- tain the Society’s excellent reputation. If elected, I will offer even entific and professional organizations associated with national more support than in the past, and I ask for your vote to help me and international universities, research institutions, companies, achieve this opportunity. and governmental institutions. An extended biography and vision statement are available at Biography. Paolo Montuschi obtained an MSc in 1984 and a http://montuschi-bog-2011.polito.it. PhD in computer engineering in 1988. Since 2000, he has been

Jane Chu Prey of Virginia. She taught in the University of Virginia’s Computer Position statement. The IEEE Computer Science Department for 11 years and joined Microsoft Research Society is the world’s premier organization in 2004, where she leads the Tablet Technologies in Higher Edu- for practitioners and academics in comput- cation initiative. She is also responsible for the development and ing. We need to use this position to become implementation of Microsoft Research’s Gender Diversity and more influential outside our current com- Pipeline Strategy. munity and to provide thought leadership Prey spent two years as the Computer Science Program Man- for both industry and academia. ager at the US National Science Foundation. She has served on a With my background in academia, gov- number of advisory boards—both academic and professional, ernment, and industry, I have been fortunate to experience many including the Dean’s Advisory Board at Virginia Tech College of different environments and colleagues. I’ve learned that academia, Engineering, the Madeira School Board of Trustees, the Blue Ridge government, and industry have many overlapping hopes and Community College Advisory Board, the ACM Education Board, goals—we just don’t say them the same way or approach problems and the ACM Special Interest Group on Computer Science Educa- the same way. That’s where the lack of collaboration comes from. tion (SIGSCE) Board. The top challenge I’d like to see the Society pursue is show- Her Computer Society activities include work on the Educa- ing leadership in working with other professional and academic tional Activities Board and Computer Society representative to organizations to develop an all-up strategy to attract, retain, and the Frontiers in Education (FIE) steering committee. She served develop women and underrepresented groups into our discipline. as both secretary and chair of the FIE steering committee, as well It would be my privilege to help work on this and other as program co-chair. Prey received FIE’s Ronald J. Schmitz Award priorities. for outstanding service. She currently serves as the Computer Society representative to the Computing Research Association’s Biography. Jane Chu Prey graduated from the University of Illi- Executive Board. nois at Urbana-Champaign and received a PhD from the University

AUGUST 2010 77

R. Sampath primary focus on humanitarian technologies for global and Indian Position statement. In today’s age of rapid markets. Sampath has over 20 years’ experience in industry and and dynamic change, IEEE and the Com- technology start-ups. He is a frequent speaker at conferences and puter Society serve as beacons of knowledge has taught in the graduate program at the University of California, and adaptation. For continued relevance, Irvine. the Computer Society should focus on new Sampath has provided IEEE and Computer Society leadership globalization needs, technology conver- at the chapter, section, and regional levels in Texas, Virginia, and gence, careers, collaboration, and innovative California. His Computer Society roles have included work on the access to knowledge. Apart from member- Publications Board and the Member Development Committee. value imperatives, industry needs, and younger member segments, As the IEEE Orange County section chair, Sampath helped pio- also essential are macro causes such as innovation, education, neer two initiatives: Engineers-in-Transition (focusing on careers and humanitarian technology efforts that have the added benefit and retraining) and Project E2E: Engineers-to-Educators (transi- of mobilizing our community. tioning technology professionals into public school systems). He The Board of Governors is responsible for guidance, strat- developed these initiatives in concert with California educational egy, direction, and oversight. I offer the experience, insight, and agencies, professional organizations, universities, and corpora- cultural sensibility to provide strong leadership and to build part- tions. In 2008, he received the IEEE-USA Regional Professional nerships that get results. I intend to spearhead these initiatives that Leadership Award for California public policy initiatives in educa- truly make a difference for our members, the computing commu- tion and engineering. He has been an IEEE Senior Member since nity, and the world at large. With your support, I hope to have the 2004. opportunity to serve you as a member of the Board of Governors. Sampath received a bachelor’s degree in electronics and communication engineering from the National Institute of Tech- Biography. R. Sampath is a director at Quanta Consulting, a Los nology, Trichy, India, and an MBA from the University of Southern Angeles-based advisory firm helping companies innovate—with a California.

Charlene (Chuck) Walrad Biography. Charlene (“Chuck”) Walrad has more than 30 years Position statement. The Computer Soci- of experience in the software and IT industries, leading the devel- ety, like our industry, is “living in interesting opment of more than two dozen commercial software products times.” Such challenges have always been across a variety of applications, ranging from the widely used a call to action and innovation for those in Systran natural language translation systems, to CA Technologies’ technology, leading to new and better solu- Ingres RDBMS system. In 1987, she founded Davenport Consulting tions and products. (www.daven.com), whose clients include companies like Adobe, Operationally, we need to continue to IBM, Microsoft, AAA, and Ford. She has spoken at numerous indus- look at how we do business and how we can try conferences and written several publications in the field. do it better, using the economies of scale that are already available. Walrad is a senior member of the IEEE and a member of the We need to attract a new generation of volunteers and listen to Computer Society’s Digital Library Operations Committee, Stan- them. Their fresh insights and can-do attitudes are energizing and dards Activities Board, and Professional Activities Board. She often question the traditional ways of doing things. serves on the Executive Committees of IEEE’s Software and Sys- We must leverage the IEEE’s credibility and brand-name recog- tems Engineering Standards Committee (S2ESC) and the IEEE nition to increase the Computer Society’s visibility and influence Computer Society Technical and Conference Activities Board, in our industry. We must preserve the Society’s long-standing and she is the working group chair for the development of a new tradition of technical excellence in the academic and research software configuration management standard. communities and actively promote these strengths in the work Honors include Marquis Who’s Who in the West, Marquis and market places that computing professionals inhabit. Who’s Who in Science & Engineering, International Who’s Who in Information and Technology, Who’s Who in Colleges and Universi- ties, Who’s Who of American Women, and Mensa. Walrad received an MS in linguistics from the University of California, San Diego.

IEEE COMPUTER SOCIETY ELECTION

Cast your vote quickly and easily on the Web at https://www.directvote.net/ieeecs______or via fax to +1 952 974 5110. Ballots must be received no later than 12:00 noon EDT on Monday, 4 October 2010. To vote by mail, use the return-mail envelope provided and send your ballot to the address provided below.

Return ballots by mail to: IEEE Computer Society c/o Survey & Ballot Systems PO Box 46430 Eden Prairie, MN 55344, USA

78 COMPUTER

REPORT TO MEMBERS

IEEE President-Elect Candidates Address Computer Society Concerns

he IEEE Computer Soci- Society members’ votes to infiuence positions synthesize the views of our ety has established a the selection of the IEEE leadership, most senior leadership: the Society’s reputation for excel- we posed questions to this year’s IEEE current, past, and incoming presi- T lence within the fields president-elect candidates. Because dents. We present these questions of computing. As a component of this election determines who will and answers (limited to 150 words the IEEE, the Computer Society’s serve as president-elect in 2011, each) to help you make your deci- activities parallel those of 37 other president in 2012, and past president sion in the IEEE annual election. IEEE societies and councils serving in 2013—vital positions within the Only ballots received by noon, Cen- the computing and engineering dis- IEEE’s governing body—our members tral Time, on 1 October 2010 will be ciplines. As the largest IEEE society must cast informed votes. counted. contingent, the Computer Society Our volunteer leaders have iden- We also remind and encourage has 85,000 members, approximately tified the following questions as you to cast your vote for Computer 60 percent of whom are full IEEE essential to the Computer Society, Society leaders by noon Eastern members. the IEEE, and the Computer Society’s Time, 4 October 2010 in our Society Recognizing the impact of the IEEE relationship with the IEEE. The rst election. leadership over the Computer Society response to each question states the —Sorel Reisman, IEEE Computer and in turn the power of Computer Computer Society’s position. These Society president-elect

GORDON W. DAY JOE LILLIE Gordon Day spent most of his Joe Lillie received a BS in elec- career in research and manage- trical engineering and an MS in ment at the National Institute of telecommunications from the Standards and Technology, where University of Southwestern Loui- he founded and led the NIST Opto- siana. He has 36-years experience electronics Division. His personal in telecommunications manage- research ranged from fundamental ment. Lillie was employed by optical measurements to the development of standards BellSouth from 1973 to 2002, retiring as a member for optical ber and new concepts in instrumentation. of the Louisiana BellSouth State Staff. In 2003, Lillie More recently, he served as science advisor to US Sena- joined NorthStar Communications Group, where he tor Jay Rockefeller and director of government relations served as the Director of Corporate Quality. In Sep- for the Optoelectronics Industry Development Associa- tember 2005, Lillie returned to BellSouth, working on tion. He has been a professor adjoint at the University Hurricane Katrina restoration. Lillie continues to proof Colorado and a visiting fellow at the University of vide engineering support to AT&T. Southampton (UK), and has served on many industry, During his professional career, Lillie has attended government, and academic advisory groups. He is a numerous training sessions on telephony, manage- past president of the IEEE Photonics Society and of ment, leadership contract administration, and quality IEEE-USA, and is a fellow of IEEE, AAAS, the Optical management. Lillie and his wife, Debbie, have been Society of America, and the Institute of Physics (UK). He married for 37 years. They have two children and ve received a BS, a MS, and a PhD in electrical engineering grandchildren. from the University of Illinois.

REPORT TO MEMBERS

QUESTION 1: COMPUTER SOCIETY POSITION

What is the highest priority that you No IEEE Computer Society position. have for the Computer Society in 1 your presidential year, and how do 1 you intend to help us accomplish it?

Day Lillie The success of the Computer Society should be a high I am in no position to set priorities for the Computer 1priority for all IEEE leaders. As a former president of 1Society; this is a task for your society. Once the Com- the IEEE Photonics Society, I understand the difculties of puter Society identifies its highest priority, I will fully providing publications, conferences, and other services in support this priority just as I have done in the past. An a competitive and budget-limited environment. Over the example is the membership initiative project which was past few years, the Computer Society has made signicant proposed by the Computer Society. I fully supported this progress in improving its business model, and I will work initiative as a member of the IEEE Finance Committee to help it continue on its current path of development. and the IEEE Board of Directors. So, my highest priority will be to listen to your issues, understand these issues, and work hard in concert with the Computer Society to address these issues.

QUESTION 2: COMPUTER SOCIETY POSITION

The IEEE Board of Directors may be The Computer Society believes that any transformation must restructured. What do you think of 2be accomplished with great caution, deliberation, in mea- the proposed changes, and how will sured steps, and with well-dened benets. Moving too quickly 2 restructuring help the Computer signicantly increases the risk of a suboptimal solution and does Society better serve its large community of not provide for adequate vetting of solutions by all stakeholders. professionals? We support transparency and accountability as this transformation takes place. We believe that outside expert experience in helping nonprots transform should be engaged. If change must occur, a clear strategy must be dened. It is essential to consider, as a priority, governance improvement and efciencies that produce a more responsive and fiexible IEEE board directed toward future growth in diversity and global membership as well as better engagement of new professionals and emerging elds.

Day Lillie I believe that IEEE members want their leaders to Recent attempts to change the composition of 2focus on improving the benets of membership and 2the IEEE Board of Directors have been unsuccess- constraining costs, rather than on governance. There ful. Changes of this type are difcult, given the process are opportunities for change that could help the Board required to implement the change. I would suggest that the function better, but I am concerned about the disruption role of the IEEE Board of Directors must be claried. The of a major reorganization. As president, I would focus Board should be focused on policy and strategic issues and the Board’s attention more on policy and strategic deci- let the organizational units handle the implementation and sions and allow the operating units to manage their own operational issues. As your president, I will work to create activities. The Computer Society has asked that any reor- a strategic agenda for the IEEE Board that focuses on what ganization be pursued “…with great caution, deliberation, IEEE should be doing worldwide and then let the organi- in measured steps, and with well-dened benets.” I agree zational units do their job in implementing this strategy. If and note that the Board’s most recent actions are consis- we take this approach, we can be a better organization and tent with that view. better serve the worldwide technical community, including the Computer Society and all members.

80 COMPUTER

QUESTION 3 COMPUTER SOCIETY POSITION

Currently, the IEEE imposes a very We believe that problem stems from high overhead costs high $24 direct expense on each soci- 3that are distributed and passed back to the societies, with ety member, presumably in part to equivocal regard for the correctness of the assessments. There is 3 cover various IEEE expenses. Do you a reasonable contribution the Computer Society should make to plan to address this assessment and situation support the IEEE in both its mission and its day-to-day operations. and, if not, why not? However, there appear to be insufcient controls in place with respect to these growing cost factors, thus impacting the ability of the Computer Society and other IEEE operating units to serve their professional communities. Most telling is the limitation such fees put on our ability to hold down membership costs and attract a much broader membership base.

Day Lillie I have promised to work to improve the cost/benet The cost charged to the Computer Society (and all 3ratio of membership across all of IEEE. That includes 3other IEEE societies and membership entities) is a society as well as IEEE membership. The example cited is per-transaction cost determined by dividing the total cost an apportioned share of the direct cost of services provided to support members by the total number of transactions. to the Society by IEEE. I will work with volunteers and This allocation process, like many IEEE cost-allocation staff to identify opportunities to reduce both direct and algorithms, has been in use for many years. Yes, I am indirect costs, while at the same time working to increase willing to address this cost allocation. The rst step is the benets of IEEE and society membership. to evaluate the overall cost, identify ways to reduce the cost, and implement a cost-reduction plan. We should also review the algorithm to ensure that the correct allocation is used. Our members deserve the best possible service at the lowest possible cost, and if we work together this can be accomplished.

AUGUST 2010 81

REPORT TO MEMBERS

QUESTION 4 COMPUTER SOCIETY POSITION

IEEE has had some false starts and The IEEE and the IEEE Computer Society have been work- has expended many volunteer and 4ing together toward signicant strategic and cooperative IT staff resources to support various IT activities that will return major benets to both. For example, the 4 efforts during the past four years. If IEEE Computer Society and IEEE have agreed to integrate various you are elected, what will you do to ensure suc- components of the Society’s digital library with the IEEE Xplore cessful delivery and improved IT services for platform. Where appropriate, we have insourced parts of our IT our members? And if you could launch a new infrastructure to IEEE and its service providers. IT effort to support membership or volunteers, From these examples, it is clear that we have a vested interest in what would that be? the success of the IEEE’s IT operations, both in terms of reducing operations costs and improving the chances for success. For these reasons, we feel that it is essential for IEEE to demonstrate effective management and control of its development processes and successful delivery of essential member applications. Perhaps it would be useful for the IEEE to provide societies with more frequent updates on the status of development of applications that are vital to member services.

Day Lillie IEEE’s IT infrastructure must support the disparate As the largest technical professional society in the 4needs of members, volunteers, customers, staff, and 4world, our members deserve better IT support than the general public. The difficulties are significant, but they have received in the recent past. The new CTO has success is essential to our future. That success will come overall responsibility for IT functions, and the IT team must primarily through the efforts of our dedicated staff, and I clearly understand the expectations of all of our stakehold- am pleased that our executive director has accepted full ers (members, volunteers, and customers). Using these responsibility for managing those efforts. I will continue expectations and available technology and resources, we to be an advocate for providing the necessary resources must have a world-class operation that is the model for and ensuring vigorous accountability. Greater use of tech- technology organizations. From a member and volunteer nology for conducting meetings, broadcasting events, and perspective, the most important IT component is the user distributing archived resources would enable more mem- interface. Therefore, if I were to launch a new IT effort, it bers to participate in IEEE activities, reduce travel costs, would contain a user interface that is easy to understand, and increase our global integration. I will work to acceler- is easy to navigate, and retains a history of the user’s IEEE ate IEEE efforts in these areas. contacts. From a volunteer perspective, the interface would also give the volunteer access to relevant real-time data that helps the volunteers perform their required functions and respects applicable privacy rules.

82 COMPUTER

QUESTION 5 COMPUTER SOCIETY POSITION

Introduction of new society pub- The Computer Society would like to see a process that pro- lications in the IEEE is a lengthy, 5vides more support for dealing with scope issues related arduous, and burdensome process. to new publication proposals, allows nascent editorial boards 5 For example, the minimum time to develop their community, and demonstrates reader interest from concept to delivery of the first issue of a in their proposals through online publications. Currently, the new publication, whether in print or online, is process for dealing with confiicting publication scopes can easily about 18 months. What will you do to make the add 6-12 months to the development of a new periodical. It is approval process more efficient? purely adversarial, with minimal dened processes for mediat- ing disagreements. Simplied procedures are required to allow all communities to identify potential confiicts and resolve them quickly. It is essential that Publications Services and the Products Board begin to formulate and test new “outside-the-box” practices to make the process of introducing of new publications more efcient. The Computer Society is more than willing to offer ideas in this regard.

Day Lillie I am committed to maintaining the quality of IEEE The current timeline for this process is unaccept- 5publications, and to keeping them at the forefront of 5able. If we are to remain relevant and competitive, rapidly advancing technologies. I have been an editor and we must do a better job of providing timely new publica- have served on committees and boards that oversee some tions. My management experience has taught me that we of IEEE’s most respected and successful publications. The must continuously review processes, identify areas for successful launch of a new publication requires careful improvement, utilize feedback, and implement improve- planning and decisions, creation of an effective editorial ment plans. If we are not meeting the expectations of the structure, solicitation and review of manuscripts, and user community, including the authors, volunteers, and advertising to potential customers, among other activities. readers, then we risk losing our base of support. We are Minimizing the time between conception and rst issue in a competitive eld and must not allow the competition is important in a competitive environment, but I think it to manage this process better than we do. As your presi- is difcult to launch a high-quality publication in fewer dent, I will insist that the IEEE do a better job with this and than 18 months. other processes. We are the largest technical professional society in the world, but we must continuously improve to maintain this elite status.

AUGUST 2010 83

CAREER OPPORTUNITIES

PROJECT MANAGER, SENIOR SOLU- number to Hewlett-Packard Company, ate standard SW development life cycle TIONS ARCHITECT, HSBC NY, NY Utilize 19483 Pruneridge Ave., MS 4206, Cuperti- methodologies & processes. Mail resume knowl. of FIX, MQ, GLtrade, ULLINK, Java/ no, CA 95014. No phone calls please. Must to Saber Software Inc., 19483 Pruneridge C++, SQL to lead various areas of user exp be legally authorized to work in the U.S. Avenue, MS 4206, Cupertino, CA 95014. & stock exchange interface design for without sponsorship. EOE. Resume must include Ref. #SABCOLSID11, investment banking (front, middle, back full name, email address & mailing ad- office) in the area of trading, pertaining dress. No phone calls please. Must be le- to fixed income, futures and options and HP ENTERPRISE SERVICES, LLC is gally authorized to work in the U.S. with- equity development for various projects. accepting resumes for the following out sponsorship. EOE. BS/MS or For. Equ. Comp. Engin., CS or positions: INFORMATION TESTING in related, plus related exp. Resumes to HR, Springfield, OR. (Ref. #ESSPRIT1). Design, Attention PD. HSBC, 452 5th Ave, Tower develop and execute all testing-related FMC TECHNOLOGIES, INC. is seeking 12, NY, NY 10018. activities on applications, infrastructure SAP Human Resource Configuration Spe- or hardware components of IT solutions. cialist. Will analyze and understand com- Please mail resume to HP Enterprise Ser- pany’s HR system and process to config- HEWLETT-PACKARD COMPANY has an vices, LLC, 19483 Pruneridge Avenue, MS ure the SAP system and facilitate the HR opportunity for the following position in 4206, Cupertino, CA 95014. Resume must business processes. Position requires Cupertino, CA. Software Designer: Reqs: include Ref. #, full name, email address & Bachelors Degree or equivalent foreign Programming languages: Java 5, Java mailing address. No phone calls. Must be degree(s) + 5 years of experience OR Mas- Swing, J2EE (JSP, Servlets, Beans, Tags), legally authorized to work in U.S. without ters degree + 3 years of experience. Job XML, SQL, C, Java Script, Hibernate, RMI, sponsorship. EOE. location: Houston. Send resume to: FMC Python, C, UML; Database: Oracle; Dvlpmt Technologies, Inc., Attn: S. Eagar, 1777 Envmt: Eclipse, Jbuilder, Microsoft, .NET, Gears Rd., Houston, TX 77067 and refer to JUnit, Linux, Unix; Operating Systems: SABER SOFTWARE, INC. is accepting job code SB0003 when applying. Windows, Sun Solaris; Web Srvr: Apache resumes for Services Information Devel- Tomcat. Also reqs: Bach degree or foreign oper in Columbus, OH. (Ref. #SABCOL- equiv in SW Engrng, CS, Engrng or rel. & SID11). Conceptualize, design, develop, SIEMENS PLM SOFTWARE INC. has an 3 yrs exp in job offered or rel occupation. unit-test, configure, & implement por- opening in Milford, OH for Software En- List full name, address & email address tions of new or enhanced (upgrades or gineer to port & build NX Nastran on new on resume. Send resume & refer to Job# conversions) business & technical SW so- hardware platforms, OS & compilers. Re- CUPHSE2. Please send resumes with job lutions through application of appropri- quires Master’s degree & coursework in finite element methods. Send resumes to ____PLMCa- [email protected].______Job code UGS78 must be Positions at the Institute for Defense Analyses referenced in email subject line. EOE. Center for Computing Sciences The Institute for Defense Analyses Center for Computing Sciences (IDA/CCS) HEWLETT-PACKARD COMPANY is MWPSSOMRKJSVSYXWXERHMRKVIWIEVGLIVWXSEHHVIWWHMJ½GYPXGSQTYXMRKTVSFPIQWZMXEPXSXLIREXMSR´W accepting resumes for Systems/Soft- WIGYVMX]-(%''7MWERMRHITIRHIRXETTPMIHVIWIEVGLGIRXIVWTSRWSVIHF]XLI2EXMSREP7IGYVMX] ware Engineer in Roseville, CA. (Ref. %KIRG] 27% )QTLEWMWEVIEWJSV-(%''7XIGLRMGEPWXEJJMRGPYHILMKLTIVJSVQERGIGSQTYXMRK #ROSSSE11). Conduct or participate in GV]TXSKVETL]ERHRIX[SVOWIGYVMX]1IQFIVWSJXLIXIGLRMGEPWXEJJGSQIJVSQEHMZIVWIZEVMIX]SJ multidisciplinary research & collaborate FEGOKVSYRHWMRGPYHMRKGSQTYXIVWGMIRGIGSQTYXIVEVGLMXIGXYVIGSQTYXIVIPIGXVMGEPIRKMRIIVMRK MRJSVQEXMSRTVSGIWWMRKERHXLIQEXLIQEXMGEPWGMIRGIW QSWXLEZI4L(W7TIGMEPEXXIRXMSRMWTEMH with equipment designers and/or HW en- XSXLIHIWMKRTVSXSX]TMRKIZEPYEXMSRERHIJJIGXMZIYWISJRI[GSQTYXEXMSREPEPKSVMXLQWXSSPW gineers in design, devlpmt, & utilization TEVEHMKQWERHLEVH[EVIHMVIGXP]VIPIZERXXSXLI27%QMWWMSR7XEFPIJYRHMRKTVSZMHIWJSVEZMFVERX of electronic data processing sys SW. De- VIWIEVGLIRZMVSRQIRXERHEREXQSWTLIVISJMRXIPPIGXYEPMRUYMV]JVIISJEHQMRMWXVEXMZIFYVHIRW sign, develop, troubleshoot, & debug SW 8LIGIRXIVMWIUYMTTIH[MXLEZIV]PEVKIZEVMIX]SJLEVH[EVIERHWSJX[EVI8LIPEXIWXHIZIPST programs. Mail resumes with reference QIRXWMRLMKLIRHGSQTYXMRKEVILIEZMP]YWIHERHTVSNIGXWVSYXMRIP]GLEPPIRKIXLIGETEFMPMX]SJXLI number to Ref. #ROSSSE11, Hewlett-Pack- QSWXEHZERGIHEPKSVMXLQWERHEVGLMXIGXYVIW-(%''7VIWIEVGLWXEJJQIQFIVWLEZIEP[E]WFIIR ard Company, 19483 Pruneridge Avenue, EXXLIJSVIJVSRXSJGSQTYXMRKEWIZMHIRGIHF]PEWXMRKZMWMFPIGSRXVMFYXMSRWXSEVIEWEWZEVMIHEW MS 4206, Cupertino, CA 95014. No phone QYPXMXLVIEHIHEVGLMXIGXYVIW IK,SVM^SR RSZIPGSQTYXMRKW]WXIQW IK*4+%FEWIH7TPEWLERH calls please. Must be legally authorized 7TPEWL4VSGIWWMRK-R1IQSV]GLMTW HIWMKRERHMQTPIQIRXEXMSRSJSTIVEXMRKW]WXIQW IKXLI to work in the U.S. without sponsorship. 0MRY\OIVRIP ERHTVSKVEQQMRKPERKYEKIHIWMKRERHMQTPIQIRXEXMSRJSVLMKLTIVJSVQERGIGSQTYX EOE. MRKW]WXIQW IK9RMZIVWEP4EVEPPIP'ERH'MRUYIGIRXS -(%''7VIWIEVGLWXEJJ[SVOSRGSQTPI\XSTMGWSJXIRIRKEKMRKQYPXMHMWGMTPMREV]XIEQW GERHM HEXIWWLSYPHHIQSRWXVEXIHITXLMRETEVXMGYPEV½IPHEW[IPPEWEFVSEHYRHIVWXERHMRKSJGSQTYXE PEABODY ENERGY CORPORATION has XMSREPMWWYIWERHXIGLRSPSK]&IGEYWIXLITVSFPIQWSJMRXIVIWXEVIGSRXMRYEPP]IZSPZMRK-(%''7 an opening in St. Louis, MO for a SAP Ana- VIGVYMXQIRX JSGYWIW SR WIPJQSXMZEXMSRWXVIRKXL SJ FEGOKVSYRH ERH XEPIRXVEXLIV XLER WTIGM½G lyst II. Duties include providing IT sup- I\TIVXMWI port to business & implement projects, 0SGEXIHMREQSHIVRVIWIEVGLTEVOMRXLI1EV]PERHWYFYVFWSJ;EWLMRKXSR('-(%''7SJJIVWE GSQTIXMXMZIWEPEV]ERI\GIPPIRXFIRI½XWTEGOEKIERHEWYTIVMSVTVSJIWWMSREP[SVOMRKIRZMVSRQIRX enhancements, & fixes in SAP PP module; 97GMXM^IRWLMTERHE(ITEVXQIRXSJ(IJIRWI877-GPIEVERGI [MXLTSP]KVETL EVIVIUYMVIH-(% designing configuration & customiza- ''7[MPPWTSRWSVXLMWGPIEVERGIJSVXLSWIWIPIGXIH8LI-RWXMXYXIJSV(IJIRWI%REP]WIWMWTVSYHXS tion of SAP R/3 applications by articulat- FIERIUYEPSTTSVXYRMX]IQTPS]IV ing business requirements & translating 4PIEWIWIRHVIWTSRWIWSVMRUYMVMIWXS them into effective solutions; & other du- Dawn Porter ties required. Requires a MS degree & 3 Administrative Manager yrs. exp. or BS degree & 5 yrs. exp. Mail IDA Center for Computing Sciences resumes to Attn: J. Hardin, Peabody En- 17100 Science Drive ergy Corp., 701 Market Street, St. Louis, Bowie, MD 20715-4300 MO 63101; reference Job ID #: SAP Analyst [email protected] II – NA.

Faculty Position www.cit.cmu.edu______

Carnegie Mellon University is embarking on Carnegie Mellon’s College of Engineering munications, Security, Software Engineering, an exciting opportunity to transform gradu- (www.cit.cmu.edu) is consistently ranked Image and Signal Processing, Entrepreneur- ate education in East Africa by introducing amongst the top ten in the USA and the ship, and Innovation and Technology Man- new models of education, research and world. The College includes 7 academic agement, who can deliver innovative, inter- development, and the commercialization of departments granting BS, MS, and PhD disciplinary graduate programs. Candidates information and communication technolo- degrees, two graduate-only degree grant- should possess a PhD in a related discipline gies (ICT). With programs from Europe to ing departments (Information Networking and an outstanding record in research, Asia, Carnegie Mellon’s College of Engineer- Institute and Carnegie Mellon Silicon Valley teaching and leadership. Applications should ing is establishing a regional ICT Centre of Campus), several multidisciplinary research include a comprehensive resume includ- Excellence (CoE) in Kigali, Rwanda that will centers, and two Institutes (Carnegie ing a complete list of publications, a list of provide first class graduate education in a Mellon CyLab, and Institute for Complex 3-5 professional references, a statement of region of the world booming with opportuni- Engineered Systems). research and teaching interests (less than 2 ties for technology innovation. Striving to pages each), and copies of 2 research papers We expect the first class of students to be become the technology hub for East Africa, (journal or conference papers). enrolled at the CoE in Kigali starting Sum- Rwanda is investing heavily in infrastructure mer/Fall 2011. Successful applicants will Applications should be addressed to: and capacity building in the critical areas of be expected to spend up to two academic Dean, College of Engineering ICT and engineering. With a history of excel- semesters at Carnegie Mellon, Pittsburgh Carnegie Mellon University lence in higher education, Carnegie Mellon starting fall 2010 before assuming their po- 5000 Forbes Ave will address Rwanda’s and the regions’ skill sition at the CoE in Kigali in summer 2011. Office: 110 Scaife Hall needs by offering Master of Science degree Pittsburgh, PA 15213-3890 programs in Information Technology and Carnegie Mellon is seeking exceptional Electrical and Computer Engineering. candidates in the areas of Networking, Com- [email protected]

THE UNIVERSITY OF PENNSYLVANIA

8LI9RMZIVWMX]SJ4IRRW]PZERME´W(ITEVXQIRXSJ'SQTYXIVERH-RJSVQEXMSR7GMIRGIMRZMXIWET TPMGERXWJSVX[S0IGXYVIVTSWMXMSRW8LIHITEVXQIRXWIIOWMRHMZMHYEPW[MXLI\GITXMSREPTVSQMWI JSVSVETVSZIRVIGSVHSJI\GIPPIRGIMRXIEGLMRKYRHIVKVEHYEXIGSYVWIW%TTPMGERXWWLSYPHLSPHE KVEHYEXIHIKVII TVIJIVEFP]E4L( MR'SQTYXIV7GMIRGISV'SQTYXIV)RKMRIIVMRKERHLEZIE WXVSRKMRXIVIWXMRXIEGLMRK[MXLTVEGXMGEPETTPMGEXMSR (YXMIWJSVXLI½VWX0IGXYVIVTSWMXMSRMRGPYHIEHZERGIHGSQTYXIVKVETLMGWGSYVWIW[MXLMRX[S TVSKVEQWXLI&EGLIPSVSJ7GMIRGIERH)RKMRIIVMRKMR(MKMXEP1IHME(IWMKRERHXLI1EWXIVSJ7GM IRGIERH)RKMRIIVMRKMR'SQTYXIV+VETLMGWERH+EQI8IGLRSPSK] WIILXXTGKGMWYTIRRIHY The Hong Kong Polytechnic University 8LITSWMXMSRWXEVXWJanuary 1, 2011 ETTPMGEXMSRWEVIHYIF]September 15, 2010. (YXMIWJSVXLIWIGSRH0IGXYVIVTSWMXMSRMRGPYHIMRXVSHYGXSV]TVSKVEQQMRKGSYVWIWJSVQENSVW is the largest government-funded tertiary ERHRSRQENSVWERHSXLIVGSYVWIW[MXLMRXLI'SQTYXIV7GMIRGITVSKVEQ8LITSWMXMSRWXEVXW.YP] institution in Hong Kong in terms of January 15, 2011 ETTPMGEXMSRWEVIHYIF] . student number. It offers programmes at 0IGXYVIVTSWMXMSRWEVIJSVSRI]IEVVIRI[EFPIERRYEPP]YTXSXLVII]IEVWEXXLIIRHSJ[LMGL ETVSQSXMSRXS7IRMSV0IGXYVIVGERFIGSRWMHIVIH7YGGIWWJYPETTPMGERXW[MPP½RH4IRRXSFIE Doctorate, Master’s, Bachelor’s degrees WXMQYPEXMRKIRZMVSRQIRXGSRHYGMZIXSTVSJIWWMSREPKVS[XLMRFSXLXIEGLMRKERHVIWIEVGL8SETTP] and Higher Diploma levels. It has a TPIEWIGSQTPIXIXLIJSVQPSGEXIHSRXLI*EGYPX]6IGVYMXQIRX;IF7MXIEX______http://www.cis.upenn. full-time academic staff strength of edu/departmental/facultyRecruiting.shtml ______around 1,400. The total consolidated )PIGXVSRMGETTPMGEXMSRWEVIWXVSRKP]TVIJIVVIHFYXLEVHGST]ETTPMGEXMSRW MRGPYHMRKXLIREQIWSJ EXPIEWXJSYVVIJIVIRGIW QE]EPXIVREXMZIP]FIWIRXXS expenditure budget of the University is in 'LEMV*EGYPX]7IEVGL'SQQMXXII excess of HK$4 billion per year. The (ITEVXQIRXSJ'SQTYXIVERH-RJSVQEXMSR7GMIRGI University is now inviting applications and 7GLSSPSJ)RKMRIIVMRKERH%TTPMIH7GMIRGI 9RMZIVWMX]SJ4IRRW]PZERME nominations for the following post: 4LMPEHIPTLME4% %TTPMGEXMSRWWLSYPHFIVIGIMZIHF]each date listed aboveXSFIEWWYVIHJYPPGSRWMHIVEXMSR Head of Department of Computing %TTPMGEXMSRW[MPPFIEGGITXIHYRXMPTSWMXMSRWEVI½PPIH5YIWXMSRWGERFIEHHVIWWIHXS____faculty- [email protected]. Please visit the following websites for 8LI9RMZIVWMX]SJ4IRRW]PZERMEZEPYIWHMZIVWMX]ERHWIIOWXEPIRXIHWXYHIRXWJEGYPX]ERHWXEJJJVSQ more information: HMZIVWIFEGOKVSYRHW8LI9RMZIVWMX]SJ4IRRW]PZERMEHSIWRSXHMWGVMQMREXISRXLIFEWMWSJVEGIWI\ WI\YEPSVMIRXEXMSRKIRHIVMHIRXMX]VIPMKMSRGSPSVREXMSREPSVIXLRMGSVMKMREKIHMWEFMPMX]SVWXEXYW The Hong Kong Polytechnic University: EWE:MIXREQ)VE:IXIVERSVHMWEFPIHZIXIVERMRXLIEHQMRMWXVEXMSRSJIHYGEXMSREPTSPMGMIWTVSKVEQW SVEGXMZMXMIW EHQMWWMSRWTSPMGMIW WGLSPEVWLMTERHPSERE[EVHW EXLPIXMGSVSXLIV9RMZIVWMX]EHQMRMW ______http://www.polyu.edu.hk/ XIVIHTVSKVEQWSVIQTPS]QIRX The Penn CIS Faculty is sensitive to “two –body problems” and would be pleased to assist with More information on the above post: opportunities in the Philadelphia region. ______http://www.polyu.edu.hk/hro/job_external.htm

CAREER OPPORTUNITIES

SR. SYSTEMS ANALYST, Shelton, CT: nent diagrams. Test, troubleshoot, train, dardization and STP tracks integration. Act as single point of contact for onsite maint existing projects/sys. Reply to: Duties involve using Java, Java Design and off shore deve. teams. Responsible iMedX, Inc., 4 Corporate Drive, Shelton, Patterns, XML, XSL, SQL for retrieving for overall delivery of strategic initia- CT 06484. data, Web Services, Websphere, and En- tives. Sched negotiation of scope creeps, terprise JavaBeans (EJB). Direct appli- change reqs. Coord reqs analysis for cations to: M.Gagne, Mailstation F110, proposed apps. Prep use case diagrams, LEAD DESIGNER/DEVELOPER. Enfield, Massachusetts Mutual Life Insurance use case specs, domain model and reqs CT. Lead technical design, development, Company, 1295 State Street, Spring- attributes matrix. Review and clear proj and testing of systems applications solu- field, MA 01111; Please Reference Job ID: artifacts. Handle team interaction during tions. Act as lead technical resource and NC50252225. analysis, des and work involving interac- work with STP (Straight Through Process- tion diagrams, class diagrams, compo- ing) design team to ensure design stan- APPLICATION DEVELOPERS FOR RE- TIREMENT SERVICES SYSTEMS, Spring- field, MA. Multiple positions available. De- velopers with strong coding skills, often Nokia Siemens Networks US, LLC (NSN) spanning multiple technical disciplines. has the following exp./degreed open positions in Primary responsibilities include analyzing and translating complex designs into Irving, TX: unit tested and maintainable code. Du- ties involve advanced SQL with Sybase Stored Procedures, relational database, Multi-Vendor Support Engineer data base tuning, technical specification Support core & applications in Nortel platforms involving CDMA/GSM/GPRS/IMS/HLR/IP/Applications/DMS100/200/300/500 documentation, and Systems Develop- networks; end-to-end wireless network support for GSM, CDMA & UMTS; recovery specialist focusing on recovery of wireless network ment Life Cycle (SDLC). Send applications with least impact to the customer, & mentor & train engineers on wireless products for GSM/GPRS/UMTS/CDMA technologies & to: M.Gagne, Mailstation F110, Massachu- other duties/skills required. [Job ID: NSN-10TX-MVSE] setts Mutual Life Insurance Company, 1295 State Street, Springfield, MA 01111; Systems Engineer Please Reference Job ID: NC50230798. Exp. with core switch network in troubleshooting, commissioning, integration, SW upgrade & onsite technical support involving GSM, UMTS, SIP, & Access technologies; implement & troubleshoot network protocols involving standards & tools for interfaces with circuit TEST LEAD. Springfield, MA. Analyze core NEs & other duties/skills required. [Job ID: NSN-10TX-SE] user and business requirements, develop and write test plans for system enhance- RF Engineer ments, upgrades, and system problem resolutions using test design and execu- Exp. with wireless technologies to utilize RF propagation principles, cellular theory, RF engineering tools & antenna theory to design, tion and multiple testing methodologies optimize & troubleshoot GSM/GPRS/EDGE wireless networks, provide seamless call exp. to customers by reducing dropped calls, including requirement gathering. Uti- network access failures, bad voice quality & other customer impacting KPIs. [Job ID: NSN-10TX-RFE] lize Web Technologies, Visual Basic, VB Scripts, and QTP (Quick Test Professional) Core Testbed Engineer to automate regression and volume test- Support, troubleshoot, & maintain GSM & WCDMA network systems to include 2G & 3G packet core protocols; IP support involving ing and use system tools for data entry configuring Cisco routers & internal MSC/MGW LAN switches (ESA/ESB); & work with MSS/HLR & MGW network elements; work with and manipulation. Direct applications to: M.Gagne, Mailstation F110, Massa- SGSN 2G/3G. [Job ID: NSN-10TX-CTE] chusetts Mutual Life Insurance Company, 1295 State Street, Springfield, MA 01111; Systems Engineer Please Reference Job ID: NC50252229. Exp. in telecom industry; exp. with wireless circuit switched core in commissioning, integrating, troubleshooting, software upgrades, and on-site technical support; test experience with GSM/GPRS with understanding standards and tools for interfacing with wireless circuit core for GSM/WCDMA technologies with some IP technology experience; GSM/GPRS exp. & level 3 technical support SAP ABAP DEVELOPER. Springfield, experience. [Job ID: NSN-TX10-SYST] MA. Responsible for maintenance, en- hancement and production support of SAP applications utilizing Advanced Busi- Systems Engineer (Product Support Engineer) ness Application Programming (ABAP) Exp. in software architecture to include MSC server, Home Location register, IPA Multimedia Gateway, & provide Tier 3 support for language, SAP ABAP development, ABAP organizational telecom products with internal hardware/software architecture knowledge; core switch network in troubleshooting, workbench, Data Dictionary Debug, SQL commissioning, integration, software upgrade & on-site technical support to involve GSM, UMTS, & Access technologies. [Job ID: tracing tools, Maestro, and SAP Sched- NSN-TX10-PSE] uling. Direct applications to: M.Gagne, Mailstation F110, Massachusetts Mu- tual Life Insurance Company, 1295 State Mail resume to: NSN Recruiter, Street, Springfield, MA 01111. MS: 4C-1-1580 INFORMATION SYSTEMS MANAGER: 6000 Connection Dr. Miami, FL. Master’s deg, exp in systems Irving, TX 75039 analysis/tech support and exp managing IT in the transport/logistics ind w/ & note specific Job ID#. spec direction toward NVOCC/Freight forwarders req’d. Resumes to: PAS Cargo

86 COMPUTER

USA, 7850 NW 25th St, #400, Miami, FL 33122.

Modeling and Simulation Faculty Positions. The Department HEWLETT-PACKARD COMPANY has an of Modeling, Simulation and Visualization Engineering at Old Dominion opportunity for the following position in University’s Batten College of Engineering and Technology invites applications Andover, MA. System/Software Engineer: for two tenure-track faculty positions beginning January 2011. The successful Reqs exp in EJB; Hibernate and RESTful applicant will have expertise and experience in core modeling and simulation API; Noelios; Skills in Linux OS incl. NFS, (M&S) areas and a commitment to quality teaching in the department’s bachelors, CIFS, LVM, routing, etc. & packaging incl. RPM, Makefiles using C, C++, Perl & shell masters, and doctoral programs. Duties include undergraduate and graduate scripts; Solid unrstdng of Linux EXT3 teaching and development of a strong, externally-funded research program. This & EXT4 (Linux Filesystems) filesytems; is an opportunity to join and help shape the first M&S department. Preference skilled at wrtng efficient SQL for POST- will be given to applicants having experience in performing interdisciplinary GRESQL & HSQLDB database; Perl & Shell research in: (1) M&S for detection, education, and treatment of human disease script; C, C++, Java & JavaScript program- with research spanning one or more of the following areas: biomedical models, ming; Dvlp user interfaces for complex medical training devices and systems, rehabilitation engineering, medical enterprise systems; JSP, CSS, HTML, XML, robotics, and telemedicine. (2) M&S for transportation system operation and Javascript, ExtJS, AJAX & their use in GUI planning with research spanning one or more of the following areas: network design. Also reqs: Masters degree in CS, modeling, multimodal logistics, supply chain management, dynamic traffic Comp Engnrng or rel & 2 yrs exp in job assignment, flow optimization, and real-time control of transportation systems. offered or rel. List full name, address & Applicants must have a Ph.D. in an Engineering or Science discipline closely email address on resume. Send resume related to M&S. & refer to Job# ANDJLE2. Please send resumes with job number to Hewlett-Pack- Applications should include a cover letter, complete resume, statement of ard Company, 19483 Pruneridge Ave., MS teaching and research interests, and three letters of reference. All application 4206, Cupertino, CA 95014. No phone materials must be submitted via email as a single pdf document to Dr. Roland calls please. Must be legally authorized Mielke, Chair, MSVE Department, at the following email address: rmayo@_____ to work in the U.S. without sponsorship. September 1, 2010 EOE. odu.edu._____ Review of applicants will begin and continue until the positions are filled. Old Dominion University is an equal opportunity, affirmative action institution and requires compliance with the Immigration COMPUTER PROGRAMMER, NJ: Dvlp, Reform and Control Act of 1986. write & maintain d/base; Analyze user needs, dvlp & customize s/ware solutions to optimize operational efficiency; May prgm web sites & other browser-based !$6%24)3%2).&/2-!4)/.\!5'534sCOMPUTER systems. BS Comp Sci/Engg, 2yrs exp. Competence in Oracle, C++, ASP.NET, SQL Fax: +1 310 836 4067 Server, J2EE, UNIX. Send res. to: Yootech Advertiser Page Advertising Sales Representatives Email: [email protected] Associates LLC, 4400 Rte 9 S, Ste 2900, Citrix Cover 4 Freehold, NJ 07728. Carnegie Mellon University 85 Europe Recruitment: Heleen Vodegel The Hong Kong Polytechnic University 85 Mid Atlantic Phone: +44 1875 825700 IDA 84 Lisa Rinaldo Fax: +44 1875 825701 Email: impress@impressmedia.______Nokia Inc. 86 Phone: +1 732 772 0160 Fax: +1 732 772 0164 com__ SUBMISSION DETAILS: Rates Old Dominion University 87 Email: [email protected]______UMUC 26 are $445.00 per column inch New England University of Pennsylvania 85 John Restchack ($500 minimum). Eight lines Product: Classiﬁed Advertising 84-87 Phone: +1 212 419 7578 Fax: +1 212 419 7589 per column inch and aver- US East Email: [email protected]______age five typeset words per Dawn Becker Advertising Personnel Southeast Phone: +1 732 772 0160 line. Free online listing on Marion Delaney Thomas M. Flynn Fax: +1 732 772 0164 IEEE Media, Advertising Dir. Phone: +1 770 645 2944 Email: [email protected]______careers.computer.org with Phone: +1 415 863 4717 Fax: +1 770 993 4423 Email: [email protected]______US Central print ad. Send copy at least Email: ﬂynntom@mindspring.______com__ Darcy Giovingo one month prior to publica- Marian Anderson Phone: +1 847 498 4520 Sr. Advertising Coordinator Midwest/Southwest Fax: +1 847 498 5911 tion date to: Marian Ander- Phone: +1 714 821 8380 Darcy Giovingo Email: [email protected]______Fax: +1 714 821 4010 Phone: +1 847 498 4520 son, Classified Advertising, Email: [email protected]______Fax: +1 847 498 5911 US West Email: [email protected]______Lynne Stickrod Computer Magazine, 10662 Sandy Brown Phone: +1 415 931 9782 Los Vaqueros Circle, Los Sr. Business Development Mgr. Northwest/Southern CA Fax: +1 415 931 9782 Phone: +1 714 821 8380 Tim Matteson Email: [email protected]______Alamitos, CA 90720-1314; Fax: +1 714 821 4010 Phone: +1 310 836 4064 Email: [email protected]______Fax: +1 310 836 4067 Europe (714) 821-8380; fax (714) Email: [email protected]______Sven Anacker Phone: +49 202 27169 11 821-4010. Email: ma______nder- Japan Fax: +49 202 27169 20 Tim Matteson Email: sanacker@_____ [email protected].______Phone: +1 310 836 4064 intermediapartners.de______

AUGUST 2010 87

CALL AND CALENDAR

CALLS FOR ARTICLES FOR IEEE CS PUBLICATIONS IEEE Internet Computing seeks papers for a May/June 2011 issue focusing on security and privacy in social networks. Current trends in social networking indirectly require users to become system and policy administrators to protect their content. This is further complicated by the rapid growth of social networks and the continual adoption of new services on these networks. Moreover, using personal information in social networks raises entirely new privacy concerns and requires new insights into security CALENDAR 27-30 Oct: FIE 2010, Frontiers in problems. Education, Arlington, Virginia; http://____ Articles are due by 1 September. SEPTEMBER fie-conference.org Visit www.computer.org/portal/web/ 12-18 Sep: ICSM 2010, Int’l Conf. on

computingnow/iccfp3______to view the Software Maintenance, Timisoara, NOVEMBER complete call for papers. Romania; http://icsm2010.upt.ro 1-3 Nov: SRDS 2010, IEEE Int’l Symp. on Reliable Distributed Systems, IEEE Software seeks papers for 20-24 Sep: Cluster 2010, IEEE Int’l New Delhi; www.scs.ryerson.ca/ a May/June 2011 issue tentatively Conf. on Cluster Computing, Herak- ______iwoungan/SRDS2010 titled “Software Components beyond lion, Crete; www.cluster2010.org Programming—from Routines to 1-4 Nov: ITC 2010, Int’l Test Conf., Services.” 22-24 Sep: ICSC 2010, IEEE Int’l Austin, Texas; www.itctestweek.org Components have been known Conf. on Semantic Computing, Pitts- in software engineering since 1968 burgh; www.ieee-icsc.org 8-12 Nov: MASS 2010, Int’l Conf. on when Doug McIlroy introduced them Mobile Ad-Hoc and Sensor Systems,

in his seminal paper “Mass-Pro- OCTOBER San Francisco; ______https://mass2010.soe. duced Software Components” (NATO 3-8 Oct: MODELS 2010, IEEE Model- ______ucsc.edu Science Committee). Since then, soft- Driven Eng. Languages and Systems, ware components have had several Oslo; http://models2010.ifi.uio.no 10-12 Nov: ICEBE 2010, IEEE Int’l revivals. Conf. on e-Business Eng., Shanghai; The guest editors welcome case 20-22 Oct: CW 2010, Int’l Conf. on http://conferences.computer.org/icebe studies, lessons learned, and success Cyberworlds, Singapore; www3.ntu.______

and failure stories in component ______edu.sg/sce/cw2010 13-19 Nov: SC 2010, Int’l Conf. for software. High-Performance Computing, Net- Articles are due by 1 November. 23-26 Oct: FOCS 2010, Symp. on working, Storage, and Analysis, New Visit www.computer.org/portal/web/ Foundations of Computer Science, Orleans; www.sc-conference.org computingnow/swcfp3______to view the Las Vegas; http://theory.stanford.edu/ complete call for papers. ______focs2010 DECEMBER 1-4 Dec: ATS 2010, Asian Test Symp., Shanghai; http://ats10.shu.edu.cn SUBMISSION INSTRUCTIONS 4-8 Dec: Micro-43, 43rd IEEE/ The Call and Calendar section lists conferences, symposia, and workshops that the IEEE ACM Int’l Symp. on Microarchitec- Computer Society sponsors or cooperates in presenting. ture, Atlanta; www.microarch.org/ Visit www.computer.org/conferences for instructions on how to submit conference ______micro43 or call listings as well as a more complete listing of upcoming computer-related conferences. 6-10 Dec: APSCC 2010, IEEE Asia- Pacific Services Computing Conf.,

EVENTS IN 2010 APSCC 2010

September ervices computing is a new cross-disciplinary čeld that covers the science and 12-18 ...... ICSM 2010 S technology needed to bridge the gap between business services and IT/ 20-24 ...... Cluster 2010 22-24 ...... ICSC 2010 telecommunication services. The 2010 IEEE Asia-Pacific Services Computing Conference is an important forum October for researchers and industry practitioners to exchange information regarding advance- 3-8 ...... MODELS 2010 20-22 ...... CW 2010 ments in the state of the art and practice of IT/telecommunication-driven business 23-26 ...... FOCS 2010 services and application services, as well as to identify emerging research topics and 27-30 ...... FIE 2010 define the future directions of services computing. November APSCC 2010 expects to attract outstanding researchers from all over the world and 1-3 ...... SRDS 2010 continues to establish its status as the leading conference on services computing. 1-4 ...... ITC 2010 Sponsored by the IEEE Computer Society Technical Committee on Services Comput- 8-12 ...... MASS 2010 ing, APSCC 2010 takes place 6-10 December in Hangzhou, China. Visit http://apscc2010. 10-12 ...... ICEBE 2010 13-19 ...... SC 2010 ______hdu.edu.cn/APSCC_home.html for complete conference details.

Hangzhou, China; http://apscc2010. hdu.edu.cn/APSCC_home.html______

7-10 Dec: ICPADS 2010, IEEE Int’l Conf. on Parallel and Distributed Call Systems, Shanghai; http://grid.sjtu.

______edu.cn/icpads10

15-18 Dec: ISSPIT 2010, IEEE Int’l for Articles Symp. on Signal Processing and Information Technology, Luxor, Egypt; www.isspit.org/isspit/2010 IEEE Pervasive Computing

seeks accessible, useful papers on the latest peer- 19-22 Dec: HiPC, Int’l Conf. on High Performance Computing, Goa, India; reviewed developments in pervasive, mobile, and www.hipc.org ubiquitous computing. Topics include hardware

APRIL 2011 technology, software infrastructure, real-world 10-15 Apr: InfoCom 2011, 30th IEEE Int’l Conf. on Computer Com- sensing and interaction, human-computer Further munications, Shanghai; www.____ interaction, and systems considerations, including ______ieee-infocom.org details:

pervasive@ deployment, scalability, security, and privacy. MAY 2011 16-20 May: IPDPS 2011, 25th IEEE computer.org Author guidelines: Int’l Parallel & Distributed Process- www. ing Symp., Anchorage, Alaska; www.____ www.computer.org/mc/pervasive/author.htm computer. ______ipdps.org org/pervasive Join the IEEE Computer Society www.computer.org

AUGUST 2010 89

BOOKSHELF

he Five Dysfunctions of a negative publicity every time some- Team: A Leadership Fable, one researches you or your business. TPatrick Lencioni. In this Wild West 2.0 offers simple yet book, the author offers a leadership extraordinarily powerful ways to fable centered on teams. The pro- proactively protect your online tagonist’s CEO faces the ultimate reputation, determine the extent shapes for different tasks. Thus, a leadership crisis: uniting a team in of reputation damage and identify self-reconfigurable robot can first such disarray it threatens to bring its original source, offset even the assume the shape of a rolling track down the entire company. Through- most savage attacks, and control to cover distance quickly, then the out the story, the author reveals how search engines rank and dis- shape of a snake to explore a narrow the five dysfunctions that go to the play results about your business and space, and finally the shape of a very heart of why teams, even the name. hexapod to carry an artifact back to best ones, often struggle. Will she Amacom; www.amacombooks. the starting point. This book collects succeed? Will she be fired? Will the org;__ 0-8144-1509-2; 264 pp. and synthesizes significant progress company fail? from more than 20 years of existing The author’s narrative reminds us nteractive Data Visualization: research previously available only in that leadership requires courage as IFoundations, Techniques, and widely scattered individual papers, much as insight. He outlines a pow- Applications, Matthew Ward, Georges thus offering an accessible guide to erful model and actionable steps Grinstein, and Daniel Keim. This the latest information on self-recon- that can be used to overcome these book provides the theory, practi- figurable robots for researchers and common hurdles and build a cohe- cal details, and tools necessary for students interested in the field. sive, effective team. This compelling representing data, information, The authors focus on conveying fable with a deceptively simple yet and knowledge in a visual form to the intuition behind the design and powerful message will speak to all support the tasks of exploration, control of self-reconfigurable robots who strive to be exceptional team confirmation, presentation, and rather than technical details. Sugges- leaders. understanding. tions for further reading refer readers Wiley; www.wiley.com; 978-0- Sample programs provide starting to the underlying sources of techni- 7879-6075-9; 230 pp. points for building one’s own visu- cal information. The book includes alization tools. Numerous datasets descriptions of existing robots and a ild West 2.0: How to Protect have been made available that high- brief history of the field; discussion Wand Restore Your Reputa- light different application areas and of module design considerations, tion on the Untamed Social Frontier, let readers evaluate the strengths and including module geometry, con- Michael Fertik and David Thompson. weaknesses of different visualization nector design, and computing and The Internet is like the Old West—a methods. The book concludes with an communication infrastructure; an frontier rich with opportunity and examination of several existing visu- in-depth presentation of strategies for hope, but also a rough-and-tumble alization systems and projections on controlling self-reconfiguration and land of questionable characters, the future of the field. locomotion; and exploration of future dubious legal jurisdictions, and A.K. Peters, Ltd.; www.akpeters. research challenges. hidden dangers. And just like the Old com;___ 978-1-56881-473-5; 513 pp. MIT Press; www.mitpress.mit.edu; West, if you want to stake out your 978-0-262-01371-0; 224 pp. territory, you have to get there first elf-Reconfigurable Robots, An and fend for yourself. SIntroduction, Kasper Stoy, David Send book announcements to On the Web, this means defend- Brandt, and David J. Christensen. [email protected]. ing your good name and reputation Self-reconfigurable robots are con- before the attacks start. Despite the structed of robotic modules that can excellent product or service you be connected in many different ways. provide, all it takes is one unhappy These modules move in relationship customer, jealous acquaintance, or to each other, which lets the robot as 0VSFYQFSUT unsavory competitor to start the a whole change shape. This shape :PVSGVUVSF rumors flying. Before you know it, shifting makes it possible for the search engines are regenerating that robots to adapt and optimize their

SECURITY From Chaos to Collective Defense

James Bret Michael, Naval Postgraduate School Eneken Tikk, Cooperative Cyber Defence Centre of Excellence Peter Wahlgren, Stockholm University Thomas C. Wingfield, George C. Marshall European Center for Security Studies

Deterrence, civil defense, collective defense, and arms control were key national security doctrines in the 20th century, and they are being reevaluated now for application to cyberspace.

he 1 July 2010 issue of tanks, and operational, which pro- the integral role information and The Economist covered scribes certain activities such as communication technology (ICT) several of the global manufacturing chemical weapons or play in most aspects of private and T information grid’s weaponizing biological agents. Struc- public affairs. Actions against this systemic weaknesses that were tural arms control in cyberspace is critical infrastructure can be crimi- uncommon knowledge to the public. difcult if not impossible, given the nal, requiring a law enforcement Among other points, the article stated nature of cyber weapons, the impor- response; involve espionage, which that “More than nine-tenths of inter- tance to nation-states of maintaining demands action by a country’s intel- net traffic travels through undersea their access to cyberspace for all ligence community; or even come fibre-optic cables, and these are aspects of warghting, and nonstate in the form of an armed attack that dangerously bunched up in a few actors’ participation in cyber-based permits military self-defense by a choke-points,” and “Internet traf- warfare and crimes. Operational nation’s armed forces. fic is directed by just 13 clusters of arms control offers more promise— This new arena has understand- potentially vulnerable domain-name particularly when combined with ably produced many historical servers” (“War in the Fifth Domain: complementary approaches such as analogies, none of which are sufcient Are Mouse and Keyboard New Weap- deterrence through preparedness to address the novel circumstances ons of Confiict?” 3 July 2010, pp. 25-26, and collective defense. of cyberspace. The nuclear analogy 28). A companion piece argues for Structural arms control is imprac- presents the most serious case. At arms control to reduce the danger to tical and difficult to enforce in the 1955 Geneva Summit, US Presi- these weak links (“Cyberwar: It Is Time cyberspace. Instead, we suggest that dent Dwight Eisenhower stated that for Countries to Start Talking about progress be made to curb aggression the “nuclear genie” is forever out Arms Control on the Internet,” The and national security threats in the of the bottle. It became apparent Economist, 3 July 2010, pp. 11-12). This cyber domain by elaborating a strat- to his successors that “nuclear dis- refiects a larger movement in legal and egy combining proactive technical armament is no longer a technical policy circles, most notably among the and legal means with operational possibility” (Stanford Arms Control Russians, to reconsider arms control arms control and deterrence. Group, International Arms Control: methodologies pioneered during the Issues and Agreements, J.H. Barton Cold War and determine if they can be PROTECTING ACCESS and L.D. Weiler, eds., Stanford Uni- applied to “the fth domain.” TO CYBERSPACE versity Press, 1976, p. 103). Although There are essentially two types Cyberspace access has been the cyber equivalent has also escaped of arms control: structural, which elevated to a national security con- the bottle, it isn’t possible to put the limits things such as missiles and cern in most nations because of genie back in through disarmament,

SECURITY

nonproliferation treaties, or other Age of Uncertainty, Kluwer Academic just-in-time from artifacts that appear forms of structural arms control. Publishers, 2003, pp. 483-488). otherwise benign. Unlike weapons of mass destruc- Decoys can be programmed with a Furthermore, in contrast to the tion, cyber weapons are an integral spectrum of options for taking action case of nuclear and other types part of the commander’s arsenal in and providing anticipatory exception of kinetic weapons, there are no conducting force-on-force and asym- handling. We can develop policy that commonly agreed-upon metrics metric warfare and will be used places boundaries on the extent and for reporting the yields of a cyber in concert with kinetic weapons to type of deception employed but pro- weapon. Metrics must account for soften up the adversary’s defenses. vide latitude for cyber operators to more than just first-order effects. According to the 2010 US Department inject creativity into deceptions to Cyber test ranges aside, software of Defense’s Quadrennial Defense increase the likelihood they will be behaves according to the environ- Review Report (Washington, D.C., effective; the boundaries delineate ment in which it executes, that being February 2010, pp. 91-92): thresholds that, if breached, would the Internet. Testing cyber weapons to result in misuse or unlawful use of determine their yields would be dif- A failure by the Department to secure decoys. cult for many reasons, such as legal its systems in cyberspace would pose Virtual inspections for cyber weap- and policy constraints on performing a fundamental risk to our ability to ons might be possible, but it’s difcult the assessments outside of isolated accomplish defense missions today to determine if we’ve been deceived (from the Internet) environments. Use and in the future. Attacks in cyber- by an articial construct mimicking of cyber weapons of unknown pedi- space could target command and the capability to be inspected. As was gree or yield might not be a concern control systems and the cyberspace found in trying to apply structural of those actors who ignore the laws of infrastructure supporting weapons arms control for chemical weapons, information confiict, but it is of great system platforms. developing treaties for cyber weapons concern to lawful combatants. is complicated because the weapon The absence of market incentives Most of the world’s information components are broadly used in the will require a well-thought-out, top- infrastructure is owned privately and ICT industry: how they’re assembled down, government-driven initiative operated for civil use. Thus, the meth- into systems determines their use. to more quickly and efciently ll in ods used for cyber attacks on national Moreover, defining specific cyber those gaps that markets cannot. Tai- security are essentially the same as sites for inspection is difficult: the loring cooperative initiatives directly those used by cyber criminals, with horizontal diffusion of cyber technol- around these areas of market failure, the purpose and use of particular ogy and potential weapons capability or at least lack of market interest, assets being the key distinguishing makes inspection several orders of would ensure that governments and factors. In particular, developing a magnitude more complex than for markets each do what’s best in the structural arms control treaty deal- kinetic weapons. cyber realm. ing with cyber weapons would be We also need to ask, “Which To the extent that Russia is not legally and technologically chal- ‘weapons’ should be controlled?” The reassured by closer cyber cooperation lenging. International law typically useful life of cyber weapons is short among NATO members, collective lags years behind the introduction relative to that of kinetic weapons, security must reach beyond NATO’s of new weapons. In cyberspace, new due in part to technology churn. Any borders to avoid the appearance or weapons—including new types of treaty specic enough to be enforce- reality of a zero-sum confrontation. weapons—appear all the time. able against a particular threat would There may be room for a much larger A further complication is that ICT be obsolete long before deployment. conception of collective security that can be dual-purpose, making it chal- Furthermore, there’s no need to stock- includes non-NATO countries such as lenging to denitively determine that pile cyber weapons because they are India, Japan, and South Korea. an ICT artifact, such as a cloud com- readily replicated and distributed. No policy works if it’s orthogonal puting service, is a cyber weapon. It’s also hard to determine whether to persistent incentive structures. To In addition, some ICT artifacts all copies of a weapon have been build a truly effective international have been designed to be lawful, as destroyed. Obfuscation techniques, cyber security structure, we need a was shown for so-called “software such as polymorphism, encryption, depth of initiative beyond a stereotyp- decoys” that serve as an airlock steganography, and malware embed- ical operations center that can detect between technology and the law (J.B. ded into rmware or hardware only and avert a hypothetical cyber attack. Michael and T.C. Wingeld, “Lawful add to the complexity of detecting Collective security in the real world Cyber Decoy Policy,” D. Gritzalis et the presence of cyber weapons. Auto- would involve a longer time frame al., eds., Security and Privacy in the mated tools could assemble weapons and such components as regulatory

92 COMPUTER

adjustments, product inspections, in a clearer set of defensive options. ible,” such that it serves as a deterrent and a rst-class training program to Today, the three criteria for active against malefactors. instantiate best practices in the oper- defense in peacetime are necessity Preparedness and collective ational community. However, efforts (exhausting all peaceful alterna- defense are viable forms of denial such as government-based product tives with a reasonable prospect of deterrence in cyberspace for all inspections are challenging to imple- success), proportionality (doing no types of actors. Just as a defender ment and make effective (B. Michael, more damage than must be done can use strong encryption to deny an “Are Governments up to the Task?” to neutralize the immediate threat adversary the ability to gather infor- IEEE Security & Privacy, vol. 6, no. 6, and prevent it from re-engaging), mation to plan an effective attack, 2008, pp. 4-5). and imminency (ensuring the enemy being prepared across legal, policy, is “irrevocably committed” to the and technical dimensions will act as DETERRENCE IN CYBERSPACE attack before launching a preemptive a deterrent to would-be aggressors At the force-on-force level, super- defense strike—either at the incom- (J.B. Michael et al., “Integrating Legal powers are equals because they can ing weapon, or a support network and Policy Factors in Cyberprepared- deter their counterparts from all-out necessary for its success). ness,” Computer, vol. 43, no. 4, 2010, attacks. Cyberspace has punitive Providing mutual assistance pp. 90-92). deterrence through threat retaliation. in “peacetime” works within the An attacked nation could respond framework of existing alliances COLLECTIVE DEFENSE via a combination of diplomatic and regional organizations, but Collective defense poses a chal- actions and military might beyond a few cyber powers (nations with the lenge for cyber powers, but even here purely cyber counterattack, possibly wherewithal to conduct large-scale their interests are unthreatened when involving kinetic weapons, although cyber campaigns) find themselves smaller powers combine to provide probably not more destructive weap- thus aligned. One problem is the better situational awareness against ons. Short of the Cold War’s mutual cyber force-on-force threats posed mutual threats—including some that assured destruction (MAD), however, by each of the cyber powers. In day- would otherwise use small nations’ punitive deterrence will likely be inef- to-day operations, this is largely systems to hide between attacks on fective against nonstate actors and a matter of intelligence operations the great powers. The analogy of rogue nations intent on spreading and espionage, which seek to obtain “draining the cyber swamp” to make terror at any cost. Despite undeniable information without leaving any the ungoverned spaces less safe for areas of conflict and competition, trace or alteration in the examined cyber terrorists offers a powerful superpowers thus share several system. incentive. common interests in cyberspace: a In high-end espionage, covert John Arquilla argues that through quick response to cyber attacks that operations straddle the legal world illuminating and inltrating terrorist minimizes risk of partial or full loss between intelligence collection and networks in cyberspace, “our ene- of access to cyberspace; concern military operations. Covert opera- mies … would likely fiee cyberspace, about disorder introduced by rogue tions often aim at leaving artifacts as this medium could no longer be nations and nonstate actors such as in a system, permitting anything trusted. This would unravel the ‘vir- terrorist organizations; cooperation from easy return access to remote- tual caliphate [of al Qaeda],’ impede and transparency in many areas controlled sabotage. This type of support functions and, more gener- where the interests of even antago- operation and the potential damage ally, slow down an already pretty nist powers are common; and clarity it causes to a nation’s defenses make slow operational tempo in the eld” on the international legal standards it vital that international lawyers and (J. Arquilla, Aspects of Netwar & the that apply. computer technologists collaborate to Conflict with al Qaeda, Naval Post- Countries defending themselves identify lawful thresholds for opera- graduate School, 2009). in cyberspace tend to operate in a tional action. Collective self-defense is a term legal gray area: the level of actual or Space and time compression in of art in international law. The potential damage clearly demands cyberspace can cause many simulta- mechanism of collective defense national-level attention but might not neous effects at numerous locations, derives from Article 51 of the United rise to the level of an armed attack just as many actors (such as “patriotic Nations Charter, whereby an “armed under international law. hackers”) can combine their capabili- attack” activates the “inherent right Aggressors benet from this legal ties to form mass effects at a chosen of self-defense,” either individually uncertainty. Shrinking the gray area place and time. To address these fac- or collectively. This same standard gives cyber operators and national tors, cyber powers must tailor their is applied by regional organizations, decision-makers greater condence “presence” in cyberspace to be “vis- most notably NATO. Under the Wash-

AUGUST 2010 93

SECURITY

ington Treaty, NATO countries are improving the exchange of infor- Handling threats of this complexity bound to defend each other in an mation among nations about best requires multifaceted skills based on attack. The most important prereq- practices for minimizing interactive technical and legal insights. A robust uisite for collective self-defense is the complexity and coupling in cyber cyberspace is not only an ICT phe- threshold of an armed attack: some- weapon systems. These systems are nomenon but a system of technical thing we haven’t yet witnessed with known contributors to what Charles components incorporating organi- any of the known international cyber Perrow terms “normal accidents,” zational and legal solutions, dened incidents. mishaps from unanticipated interac- and implemented proactively. Rely- As serious as the 2007 cyber tion of multiple forces in a complex ing on traditional military, legal, and attacks against Estonia were, few system (Normal Accidents: Living with technical remedies is too simplistic. argued they were the legal equiva- High-Risk Technologies, Princeton Deterrence and operational arms lent of “tanks across the border.” University Press, 1999). Complex inter- control alone will not sufce. Tanks literally crossed Georgia’s actions in this context are “those of border in 2008, so the question of unfamiliar sequences, or unplanned James Bret Michael is a professor whether the incursion’s cyber com- and unexpected sequences, and either of computer science and electrical ponents would have risen to the level not visible or not immediately com- engineering at the Naval Postgradu- of armed attack never occurred. Even prehensible.” We can envision the ate School. Contact him at bmichael@______an attack that rose to the level of possibility of an actual mishap such nps.edu._____ widespread destruction would come as an inadvertent launch of cyber Eneken Tikk is a legal advisor at the from a primarily political decision to weapons against a nonexistent threat Cooperative Cyber Defence Centre of engage in cyber war. This makes it because of false information provided Excellence located in Tallinn, Estonia. critical for nations to consider their by cyber-based sensors, which in turn Contact her at eneken.tikk@ccdcoe.______collective interests and build “cyber results in an armed confiict. Perrow org.__ coalitions” in which analysts view afrms this point: Peter Wahlgren is a professor of law threats and defense in a similar set- and information technology at Stock- ting. Gradually, understanding would The defense system is one that grows holm University. Contact him at peter. create deterrence in itself as the legal in complexity. We are not merely [email protected].______framework applied to a cyber-armed adding safety devices to buffer a fail- Thomas C. Wingfield is a professor attack would give the target nation ure … [because] the “failures” we must of international law at the George many tools and remedies compared guard against are those that an enemy C. Marshall European Center for to the gray area below the law of (supremely clever and resourceful, it is Security Studies. Contact him at armed confiict. assumed) is actively promoting. Each thomas.c.wingfield@marshallcenter.______side mindlessly makes it less possible org.__ OPERATIONAL ARMS for the other to rest assured that it does CONTROL IN CYBERSPACE not seek its total destruction. Operational arms control plays a Editor: Jeffrey Voas, National Institute role in cyberspace. According to Pál The expansion of cyberspace of Standards and Technology;

Dunay, the four discernable types renders traditional legal and policy [email protected] of operational arms control are approaches to solve problems inef- measures to: facilitate communica- fective. Similarly, the speed and tion in a crisis or, in general, reduce opacity of cyber operations, as well secrecy and increase transparency as the rapid development of subjects Disclaimer in military matters, make force pos- that might be used as cyber weapons, The views and conclusions tures less offensive, and indirectly makes strategies for relying on struc- contained herein are those of decrease the reliability of armed tural arms control inappropriate. The the authors and should not be forces and their weapons (P. Dunay, nonstate nature of many cyber actors interpreted as necessarily rep- “Arms Control in the Post-Cold War is another factor hampering such a resenting the ofcial policies or World,” P. Dunay et al., Open Skies: course of action. endorsements of their respec- A Cooperative Approach to Military tive governments. Transparency and Confidence Building, dmittedly, the legal, policy, United Nations Institute for Disarma- and technological demands ment Research, 2004, pp. 5-16). A raised here are daunting, Selected CS articles and columns Let’s consider the case of imple- but the real-world benefits would are available for free at menting transparency measures by eventually outstrip past doctrines. http://ComputingNow.computer.org.

94 COMPUTER

AI REDUX Preventing Future Oil Spills with Software- Based Event Detection

S.S. Iyengar, Supratik Mukhopadhyay, Christopher Steinmuller, and Xin Li, Louisiana State University

Complex event processing systems detect problems in mission- critical, real-time applications and generate intelligent decisions to modulate the system environment.

n 20 April 2010, the beaches and may even reach the warnings were apparently ignored in BP-owned, Transocean- Atlantic seaboard. BP’s decision-making process since the operated Deepwater The oil spill has also decimated the low probability of a disaster masked O Horizon oil-drilling rig local economy, much of which relies the risk associated with such action. exploded, killing 11 workers and injur- on fishing, tourism, and deepwater Complex event processing (CEP) ing 17, and sending massive quantities drilling (banned by the federal gov- systems detect events in mission- of crude oil riddled with lethal toxins ernment in the wake of the spill), and critical, real-time applications and from the sea floor into the Gulf of the ultimate cleanup, reclamation, generate intelligent decisions to Mexico. Geophysicists, Earth-space and litigation costs will be in the bil- modulate the system environment scientists, and policymakers continue lions of dollars. (D. Luckham, The Power of Events: An to debate the cause of the explosion, In short, the Deepwater Horizon oil Introduction to Complex Event Process- how much oil has been released, how spill is one of the largest and costliest ing in Distributed Enterprise Systems, best to contain it, and the long-term in history, with far-reaching effects Addison-Wesley, 2003). In the case of impact. on the Gulf Coast that will be felt for deepwater oil drilling, advanced CEP As BP made one futile attempt decades to come. technology could have helped to pre- after another to cap the well, hun- vent the current crisis in the Gulf of dreds of millions of gallons poured COMPLEX EVENT Mexico. into the gulf, eclipsing the 1989 PROCESSING Exxon Valdez spill in less than a While BP’s “milestone maneuver” CIM SHELL week and causing incalculable envi- on 19 July to stop the flow of oil may At LSU, we’ve developed the Cogni- ronmental damage. More than 2,500 ultimately succeed, it begs the ques- tive Information Management Shell animals have already been killed, tion: could this awful tragedy have (CIM Shell), a CEP system that can including hundreds of endangered possibly been prevented in the first analyze complex events and activi- sea turtles, and the marshes and place? Instead of the chaos occurring ties and adapt rapidly to evolving wetlands that protect Louisiana’s today, could the company simply situations in a wide variety of envi- coast from erosion are seriously have ordered a maintenance crew to ronments (www.2theadvocate.com/ threatened. Scientists also fear that a the rig or perhaps shut it down? ______news/business/93725659.html).By hurricane could churn the oil-soaked BP has released documents indi- archiving past events and cross- waters, endangering many inland cating that it had concerns about the referencing them with current areas and landmarks. And should rig as far back as mid-2009, as well as events, the system can discover deep the oil reach the Gulf Stream loop numerous warnings that something patterns and then act upon them. current, it will wash up on Florida’s was amiss prior to the blowout. These Agent-based techniques continu-

AI REDUX

text, video, and so on—to provide a Bootstrap the system first: -Define the goal—for example, locate potential oil leaks. uniform basis for decision-making. -Provide initial set of trend-detection “rules”—for example, An interactive/intelligent sensing, look where oil has leaked before. inspection, and visualization system provides real-time monitoring and 1. Multiple potentially widespread analysis. sensors continually detect and CEP systems can reason about store events for cross-referencing causality, knowledge, belief, risk, and with a history of past events. uncertainty (P. Singh et al., “A Risk Reduction Framework for Dynamic 6. New information that human 2. Predefined rules applied operators become aware of can to incoming events trigger Workflows,” Proc. 2008 Int’l Conf. easily be “injected” into the detection of patterns or trends Services Computing, vol. 1, IEEE CS running system to further of significance in real time. Press, 2008, pp. 381-388) and are able tweak and augment the rules, Human operators adding to likelihood of “tweak” system to to make decisions in the presence of achieving goal. better achieve goal. incomplete information. CIM Shell 3. System executes an action or multiple actions in pursuit can determine how effective the ini- of goal. tial seed of information is and, along 5. AI techniques autonomously with a human operator, modify the and continually adjust system conditions as warranted. Even rare parameters in real time to continually enhance events with low probability can trig- achievement of goal. 4. System analyzes ger a response if the associated risk is success in achieving high; the system will magnify the risk goal. associated with an event related to a mission-critical scenario.

Figure 1. High-level view of Cognitive Information Management Shell. OIL SPILL PREVENTION Figure 2 shows how CIM Shell ally adjust CIM Shell’s parameters in eventual consistency over the net- could have prevented an oil spill near real time to adapt the system to work. Crawlers running 24/7 on the like that in the Gulf of Mexico. Had changing environments, and human cloud infrastructure index video and the system been integrated with the operators can easily add information other types of data offline. A variation Deepwater Horizon infrastructure, it to tweak the system, making goals of Friedemann Mattern’s global vir- would have interpreted prior mainte- easier to achieve. tual time (GVT) algorithm (F. Mattern, nance incidents and concerns voiced Figure 1 shows a high-level view “Efficient Algorithms for Distributed by BP’s engineers as a complex event of CIM Shell. We’ve used the system Snapshots and Global Virtual Time and magnified the associated risks. in numerous scenarios, including soil Approximation,” J. Parallel and Dis- It would have advised remote con- and water management and video tributed Computing, Aug. 1993, pp. sole operators to shut down the rig control in the presence of frame 423-434) allows for uniform inter- for maintenance or could even have losses. Deployed over a hardware pretation of temporal events over a taken the action itself autonomously. cloud infrastructure with GPU accel- wide-area network. In a non-real-time situation, we erators, it can handle around 100,000 Distributed intelligent agents tend to reason well about risks asso- events and make a million inferences (communicating state machines) ciated with a decision—for example, every second. can be created manually or gener- even though there’s a very small Using a distributed storage schema ated automatically in a model-driven chance of getting sick while travel- and middleware, CIM Shell provides way and dynamically deployed ing abroad, most of us would still persistent notification of events on over the network to meet mission buy travel insurance because the top of reliable churn-tolerant group requirements (R. Bharadwaj and S. cost associated with such an illness communication. The storage schema Mukhopadhyay, “A Formal Approach could be enormous. The Deepwater archives events as key-value pairs. to Developing Reliable Event-Driven Horizon disaster underscores the fact An implementation of the Paxos Service-Oriented Systems,” Proc. that sometimes events happen too algorithm (L. Lamport and M. Massa, 32nd Ann. IEEE Int’l Computer Soft- fast for humans to cope with. It also “Cheap Paxos,” Proc. 2004 Int’l Conf. ware and Applications Conf., IEEE CS shows that human decision making Dependable Systems and Networks, Press, 2008, pp. 227-230). The agents based on sensed data often isn’t ratio- IEEE CS Press, pp. 307-314) ensures can fuse disparate streaming data— nal and tends to underestimate risks,

96 COMPUTER

especially if rare events are involved. Relevant parameters need In such scenarios, normative decision to be captured/measured: making may not be an appropriate Interior pressure Water temperature model. Water/coil quality Oil rig console operators are Barnacles . . . responsible for an asset worth $100 million to $1 billion, not taking into account the potential environmental and health risks of system failure; in comparison, a fighter pilot controls an asset worth between $15 million and $150 million. Each console Remote operator receives information from center about 1,500 sensors and can’t possibly think clearly about the enormous Abnormal interior pressure High temperatrure Leaking Analysis + Prediction risks associated with responding or Figure 2. Had CIM Shell been integrated with the Deepwater Horizon infrastructure, not responding to events in such a it would have interpreted prior maintenance incidents and concerns voiced by BP’s data stream. engineers as a complex event and magnified the associated risks. CEP engines like CIM Shell that sense events and respond based on provable decision-making theories inventory after too many seem- S.S. Iyengar is Roy Paul Daniels can mitigate the problems arising ingly unrelated disturbances in Professor and chairman of the Depart- from such limitations in human rea- a warehouse. ment of Computer Science, as well as soning. They can help ensure safe t Factories that detect too many director of the Robotics Research Lab- oratory, at Louisiana State University. operation in the petrochemical indus- defective parts could order the Contact him at [email protected].______try much like autopilot technology parts inspected and switch to a significantly contributes to safety in different stockpile. Supratik Mukhopadhyay is an assis- the aviation industry. t A fast-food restaurant manager tant professor in the Department of could check consumption of Computer Science at Louisiana State University. Contact him at supratik@ OTHER APPLICATIONS each food item on the menu and ______csc.lsu.edu. The practical uses of distributed stock the freezer accordingly. event detection and monitoring and Christopher Steinmuller is an under- systems like CIM Shell are enormous, CIM Shell can analyze and then act graduate student in the Department of ranging from enterprise management upon nearly any conceivable complex Computer Science and in the Depart- ment of Physics and Astronomy at to golf-course sprinklers to a hospi- activity. Using distributed databases Louisiana State University. Contact tal patient’s intravenous pump. Some and capable of operating over a wire- him at [email protected].______examples: less network, the system is so intuitive and fault-tolerant that any type of Xin Li is an assistant professor in the t Drone ships, robotic freight- user could create new rules or goals Department of Electrical and Com- puter Engineering and at the Center ers with a small human crew with the simple push of a button. One for Computation & Technology at that are capable of docking person could run an entire operation Louisiana State University. Contact themselves and avoiding ocean alone, or the system could operate him at [email protected].______hazards, could revolutionize the autonomously from human input save shipping industry. for occasional inspection. We thank Daniel Walker of Praeses Corp. for helping to clarify the con- t Advanced credit-monitoring ceptual view of CIM Shell from the systems that better detect anom- ith CEP technology, BP application side and Ramesh Bha- alies in purchase activity could likely could have pre- radwaj of the US Naval Research help prevent identity theft. Wvented the Deepwater Laboratory for actively collaborating t Power-conservation systems Horizon oil spill. Indeed, CIM Shell with us on this project. could remotely shut down and similar systems being developed machines after long periods of like Apama (http://web.progress.com/ Editor: Naren Ramakrishnan, Dept. of inactivity or if no one is there to en/apama/index.html)______may make Computer Science, Virginia Tech, Blacksburg, operate them. such disasters mere memories in the VA; [email protected] t A security system could order an future.

AUGUST 2010 97

EDUCATION Computer Science: Is It Really the Scientific Foundation for Software Engineering?

Stephen T. Frezza, Gannon University

There are significant differences between an engineering discipline founded on a natural science and one founded on a formal science.

oftware engineering is both the intangible nature of software language, reviewed it, and use it to often referred to as an and to the discrete nature of software develop undergraduate software “emerging” discipline— operation. It seeks to integrate the prin- engineering programs around the S something that’s relatively ciples of mathematics and computer world have found this relationship new and for which educational stan- science with the engineering practices useful. One particular point of util- dards and definitions are needed. developed to produce tangible, physical ity was in helping fellow computing Fortunately, volunteers from the IEEE artifacts.” faculty understand the similarities Computer Society and the ACM took and differences between computer on this task, with the result being This definition, designed to science and software engineering. the Software Engineering 2004 Cur- support the development of under- That is, until the practical question riculum Guidelines for Undergraduate graduate programs, emphasizes of engineering accreditation became Degree Programs in Software Engi- the importance of the relationship a reality. neering (http://sites.computer.org/ between software engineering and For all undergraduate engineering ccse/SE2004Volume.pdf).______According computer science. The “important programs in the US, the organization to SE04: aspect,” that software engineer- responsible for an undergraduate ing builds on computer science and engineering program’s accreditation “Software engineering is that form of mathematics, stands out as central to is ABET’s Engineering Accreditation engineering that applies the principles the definition of the discipline. Commission. The EAC defines nine of computer science and mathematics general criteria that must be met for to achieving cost-effective solutions to BASIC SCIENCE VS. the program to include the title “engi- software problems. . . . One particularly ENGINEERING SCIENCE neering” (www.abet.org/Linked%20 important aspect is that software engi- From my own experience, I would ______Documents-UPDATE/Criteria%20

neering builds on computer science and say that software engineering builds ______and%20PP/E001%2009-10%20 mathematics. But, in the engineering on computer science and mathemat- ______EAC%20Criteria%2012-01-08.pdf). tradition, it goes beyond this technical ics in the same way that my own EAC Criterion 5 specifies the basis to draw upon a broader range degrees in electrical engineering minimum expectations for an engi- of disciplines. . . . Software engineer- build on physics and mathemat- neering program’s required courses: ing . . . is different in character from ics. Clearly, the professionals at the “one year of a combination of college other engineering disciplines, due to IEEE CS and ACM who drafted this level mathematics and basic sciences

Table 1. Some difierences between science and engineering in the UK. (some with experimental experience) Science Engineering appropriate to the discipline.” This is regularly interpreted to mean that 25 Goal: pursuit of knowledge and understand- Goal: creation of successful artifacts and sys- percent of the credits required of the ing for its own sake tems to meet people’s wants and needs program consist of mathematics and Key scientiffic process: discovery (mainly by Corresponding engineering process: inven- basic science courses. Historically, controlled experimentation) tion, design, production this has been interpreted as “natural Analysis, generalization, and synthesis of Analysis and synthesis of design science” and “mathematics” or other hypotheses course offerings from traditional sci- Reductionism, involving the isolation and Holism, involving the integration of many ence and mathematics departments. deffinition of distinct concepts competing demands, theories, and ideas ABET clarifies this further: “The Making more or less value-free statements Activities always value-laden engineering sciences have their roots The search for and theorizing about causes, The search for and theorizing about processes, in mathematics and basic sciences such as gravity, electromagnetism such as control, information, networking but carry knowledge further toward Pursuit of accuracy in modeling Pursuit of suficient accuracy in modeling to creative application. These studies achieve success provide a bridge between mathemat- Experimental and logical skills Design, construction, test, planning quality ics and basic sciences on the one assurance, problem solving, interpersonal hand and engineering practice on communication skills the other.” While no official interpretation build its science and mathematics the limited credit hours available for a basic science is provided, the coursework solidly around computer in the program. Consequently, any terminology is meant to distinguish science courses, particularly founda- undergraduate software engineering a physics or geology course from tional ones. program built with computer science engineering science courses such as Conversely, one could easily read as its foundation instead of natural thermodynamics or materials sci- the EAC criterion to view all computer science and traditional mathematics ence. The rationale is probably to science coursework as engineering would at best receive a “deficient” prevent undergraduate engineering science—after all, programming rating for EAC Criterion 5. Depend- programs from artificially declaring is a form of engineering, a creative ing on your view of engineering as a engineering courses as basic science endeavor that produces a real prod- discipline, this may or may not be a and unbalancing the program with uct. Computer science appears good thing. something less rigorous than is oth- significantly disconnected from the If computer science is an engi- erwise warranted. scientific method that underpins the neering science that “bridges science The open question is where com- study of natural science. From this and mathematics to engineering puter science fits in. Is it mathematics, perspective, all computer science design,” then it hardly qualifies as a basic science, or an engineering sci- courses, even foundational ones, the scientific foundation for soft- ence? Is it some mix of two or even should be classified as “engineering ware engineering or any other form all three? Here the more established science,” not “mathematics and basic of engineering, at least from the engineering definitions, as evidenced science.” ABET and similar perspectives. On by the EAC criterion, differ signifi- One would like to think that this the other hand, if it isn’t just an engi- cantly from the software engineering discrepancy is just an exercise in neering science, then its scientific definitions, as presented in SE04. pointless academic epistemology. content and methodology would need Reading the EAC criterion, the Unfortunately, it has significant rami- to be understood for well-formulated expectation might be that some, fications for engineering education, software engineering programs. perhaps a significant, amount of com- particularly computing education. Distinguishing computer science as puter science coursework would be How computer science coursework basic science, engineering science, considered as basic sciences appro- is viewed by engineers and accredit- or some blend is critical for software priate to the discipline of software ing bodies fundamentally skews how engineering as both a discipline and engineering—in particular, those undergraduate software engineering an educational program. computer science courses signifi- and other computing programs are cantly grounded in mathematics, built. COMPUTER SCIENCE IN such as data structures, algorithms, The reality is that all undergradu- ENGINEERING PROGRAMS computability, and formal methods. ate engineering programs are under In the UK education system, engi- From this perspective, an undergrad- significant pressure to maximize neering science has a more formal uate program would be expected to the material and learning within meaning and focuses on experi-

AUGUST 2010 99

EDUCATION

Table 2. Comparison of Essential Cell Biology’s table of contents with that of Data Structures Using C++.

Biology text Engineering text

1. Introduction to Cells 1. Software Engineering Principles and C++ Classes 2. Chemical Components of Cells 2. Object-Oriented Design (OOD) and C++ 3. Energy, Catalysis, and Biosynthesis 3. Pointers and Array-Based Lists 4. Protein Structure and Function 4. Standard Template Library (STL) I 5. DNA and Chromosomes 5. Linked Lists 6. DNA Replication, Repair, and Recombination 6. Recursion 7. From DNA to Protein: How Cells Read the Genome 7. Stacks 8. Control of Gene Expression 8. Queues 9. How Genes and Genomes Evolve 9. Search Algorithms 10. Manipulating Genes and Cells 10. Sorting Algorithms 11. Membrane Structure 11. Binary Trees 12. Membrane Transport 12. Graph Algorithms 13. How Cells Obtain Energy from Food 13. Standard Template Library (STL) II 14. Energy Generation in Mitochondria and Chloroplasts A. Reserved Words 15. Intracellular Compartments and Transport B. Operator Precedence 16. Cell Communication C. Character Sets 17. Cytoskeleton D. Operator Overloading 18. The Cell Division Cycle E. Header Files 19. Genetics, Meiosis, and the Molecular Basis of Heredity F. Additional C++ Topics (Inheritance, Pointers, and Virtual Functions) 20. Tissues and Cancer G. Problem Solving Using Object Oriented Methodology — H. C++ for Java Programmers — I. References for Further Study — J. Answers to Odd-Numbered Exercises

mental design and analysis. Table 1 ence textbook may help. authors begin to shift the focus to the highlights the differences. Table 2 shows a side-by-side com- application of this general knowledge Taking some clues from the UK parison of the table of contents of to practical issues, such as “Tissues perspective, a working definition Essential Cell Biology (B. Alberts et al., and Cancer.” for engineering science realizable in 3rd ed., Garland Science, 2009) with The table of contents for the data the classroom would be a bridging that of Data Structures Using C++ structures text is for a course that fol- or blend of science and engineering (D.S. Malik, 2nd ed., Course Technol- lows most introductory programming perspectives: ogy, 2009). I chose these based on courses. It’s what SE04 designates as their topics and their popularity on part of computer fundamentals and t discovery that leads to invention; Amazon.com. is required of accredited computer t hypotheses that support design; In looking at the 20 chapters of the science programs. In this text, an t isolation of specific concepts to biology text, most of these topics—at exactly opposite view is presented: support the integration of mul- least the first 18—put a pedagogical the first two chapters focus on relat- tiple ones; and focus on the transfer of knowledge ing the material to engineering and t understanding sufficiency in of cell biology as it’s currently under- design, and the following 11 chapters, model accuracy. stood, including the recognition of as well as most of the appendices, the structures involved and how focus on the transfer of knowledge The practical question to answer these structures interact to form cell of computing data structures and would be where most computer behavior. Most of the text focuses how these structures operate both science courses fall along this on terminology, definitions, and the mathematically as well as in a formal scale—science, engineering science, many important processes and rela- description language. Most of the text engineering. A cursory comparison tionships that make up cell biology. appears to focus on terminology, of a natural science to a computer sci- Only in the last two chapters do the definitions, and the many impor-

100 COMPUTER

tant processes and relationships that are data structures? How are they science. This certainly isn’t surpris- make up computing data structures. expressed? Which are more efficient ing to most software engineers. Again, you could argue that than others in different situations? what students do for homework in The application or bridging to engi- Stephen T. Frezza is an associate pro- the data structures course is sig- neering is a secondary, not a primary, fessor of software engineering in the nificantly different from that in cell focus of most coursework. From the Computer and Information Science Department at Gannon University. biology. The questions and exercises student learning perspective, data Contact him at ______frezza001@gannon. in a computer science course invari- structures aren’t significantly differ- edu.__ ably include the creation of artifacts, ent from biology—it’s really the study whereas the biology course does not. of a formal science, not a predomi- The goal is to teach students how to nantly engineering science. Editor: Ann E.K. Sobel, Department of use data structures to build useful Computer Science and Software Engineering, code; the artifacts bridge mathemat- hether you agree with Miami University; [email protected] ics and science, so this coursework this or not, expect there should be classified as engineering or, W to be significant dif- at the very least, engineering science. ferences between an engineering Selected CS articles and columns This interpretation, unfortu- discipline founded on a natural sci- are available for free at ____http:// nately, stands at odds with the IEEE ence and one founded on a formal ComputingNow.computer.org. CS and ACM assertions in SE04’s Guiding Principle 2 on computer science and its relationship to software engineering:

“Software Engineering draws its foundations from a wide variety of disciplines. Undergraduate study of software engineering relies on many areas in computer science for its theoretical and conceptual foundations, but it also requires students to utilize concepts from a variety of other fields, such as mathematics, engineering, and project management, and one or more application domains. All software engineering students must learn to integrate theory and practice, to recognize the importance of abstrac- tion and modeling, to be able to acquire special domain knowledge beyond the computing discipline for the purposes of supporting software development in specific domains of application, and to appreciate the value of good design.”

From my own perspective as a software engineering educator, I sus- pect that the IEEE CS and ACM have a LISTEN TO GRADY BOOCH point. From the students’ perspective, the data structures course is substan- “On Architecture” tively much more like math or biology than engineering. The artifacts gener- podcast available at http://computingnow.computer.org______ated aren’t about invention, design, or discovery, rather the work is primarily focused on understanding: What

AUGUST 2010 101

INDUSTRY PERSPECTIVE

IBM’s University Programs

James C. Spohrer, IBM University Programs

As IBM looks forward to the coming decade, it pauses to reflect on its history of academic partnerships.

BM has historically been asso- space missions, retail and electronic support for Eclipse development envi- ciated in student, faculty, and data exchange networks, satellite ronments, sponsorship of the ACM’s university administrators’ control systems, weather prediction, ICPC programming competition, and I minds with the core informa- and related foundational IT applica- donation of IP to open standards tion technologies needed to run the tions for business and society. groups helped faculty and students world’s most complex organizations Advances in disk drives, mag- transition innovative ideas into new and systems. In 1945, a few years netic storage systems, lithography, market offerings. before the transistor’s invention, and many other component tech- More recently, the IBM-Google members of IBM’s technical staff nology areas couldn’t have been Cloud Computing University Initia- and faculty at Columbia University’s sustained without strong academic tive partnership with the National Watson Scientific Computing Center partnerships. Science Foundation has provided were already co-teaching some of several university partners with the earliest classes in what would TOMORROW’S access to cloud computing capabili- eventually become the academic PIONEERS, TODAY ties for advanced research projects computer science discipline. An IEEE A few decades later, higher educa- such as Florida International Univer- Annals of the History of Computing tion’s perception began to evolve as sity’s Terray effort, which explores report documents these pioneering IBM’s partnerships with other com- massive analytics for geospatial decades when mainframe comput- panies brought large numbers of PCs datasets. Likewise, IBM university ing arrived on college campuses (W. to campus. This helped create the program assistance to North Caro- Aspray and O.B. Williams, “Arming first generation of computer-literate lina State University as it developed American Scientists: NSF and the business leaders in business schools. the Virtual Computing Lab (VCL) Provision of Scientific Computing Meanwhile, IBM’s researchers and software, donated to the Apache Facilities for Universities, 1950-1973,” production process specialists were Foundation, has made cloud com- IEEE Annals of the History of Comput- working with physicists, chemists, puting more accessible to educational ing, vol. 416, no. 4, 1994, pp. 60-74). and electrical engineers from uni- institutions. Again, higher educa- The IEEE report specifically high- versity research centers in the US, tion’s perception of IBM changed, lights IBM’s role in working with Europe, and Asia to improve PC com- with more emphasis on open inno- universities to co-create the work- ponents, including disk drives and vation among large corporations to force needed to advance the US and energy-efficient multicore processors. accelerate co-creation and adoption global economy, with innovations in As open innovation with uni- of new technologies (H. Chesbrough, core business accounting, the For- versities gained momentum from Open Business Models: How to Thrive tran scientific computing language, Internet and Web applications, IBM’s in the New Innovation Landscape, defense systems, airline scheduling, endorsement of Linux for e-business, Harvard Business Press, 2006).

Over the decades, students’ per- partnerships play in stimulating and tools, systems, and other content for ceptions of what it means to work sustaining regional and global eco- teaching and education at universi- for and with IBM have evolved nomic development (see E.P. Trani ties. Students can gain experience around IT for the changing needs and R.D. Holsworth, The Indispens- and certifications important to their of business and society. IBM has able University: Higher Education, transition into the workplace. For hired many thousands of university Economic Development and the Knowl- example, INNOV8 is an educational graduates and provided internship edge Economy, Rowman & Littlefield simulation in which students learn to opportunities for science, engineer- Education, 2010). balance quality and productivity as ing, management, and information Today, we measure the vital- part of a business-process manage- school students around the world. ity of university programs for IBM ment task for call centers. Universities have provided the talent and many other industry partners The IBM Cloud Academy and that keeps IBM innovative, enabling in terms of “the five pillars” or “five IBM Academic Cloud provide new the nearly century-old corporation to R’s”: research, readiness through initiatives aimed at making access continuously reinvent itself. skills, recruiting, revenue, and to a broad range of tools easier. In responsibilities. addition, the IBM Innovation Cen- THE FIVE R’S ters and University Ambassadors University partnerships continue provide regional enablement for to expand steadily. For example, Grand challenge projects universities and other IBM business IBM is working with a consortium interconnect university, partners on IBM tools and systems. of universities on the Watson deep- industry, and government Hundreds of thousands of students question-answering supercomputer benefit from these programs annu- groups. project to create the next generation ally. Rethinking skills depends on of natural language human interface maintaining an ongoing dialogue capability, which is set to revolution- among industry, academia, and pro- ize the speed of access to Web- and Research fessional associations (www.ieee. enterprise-scale information sources At IBM, four award programs focus org/education_careers/education/

(www.ibm.com/press/us/en/______on Shared University Research (SUR), ______tee_conference/index.html). IEEE and

pressrelease/27324.wss).______Open Collaborative Research (OCR), other professional associations play a Another university consortium Faculty Awards, and PhD Student Fel- key role in encouraging the dialogue is partnering with IBM on the Cogni- lowships (http://download.boulder. and stimulating needed debate (www.___ tive Computing project, which aims ibm.com/ibmdl/pub/software/dw/ computer.org/portal/web/computer/

to understand the brain’s architec- university/awards/IBM_awards.pdf).______computingpractice).______ture and provide the foundation for These highly competitive programs even more powerful, low-energy establish relationships between IBM’s Recruiting

supercomputing in the future (www.____ research, business divisions, and Working with university place- ibm.com/press/us/en/pressrelease/ universities. ment offices, IBM human resources ______28842.wss). Hundreds of award winners ben- recruiters provide information and In partnership with Texas’s Rice efit from these programs annually. assistance about both full-time and University, among others, IBM is In addition, the IBM Centers for internship career opportunities working to apply the latest advances Advanced Studies exist in partner- worldwide. in processor technology to the chal- ship with more than 40 universities The HR goal seeks to identify and lenge of biomedical supercomputing, worldwide, supporting a growing match the top students from science, including cancer research. number of IBM Research collabora- engineering, mathematics, analytics, Significantly, these and other tories in major cities. They focus on management, information, public grand challenge projects represent new knowledge creation to improve policy, humanities, and other schools public and private partnerships that IT and its applications while devel- to positions in IBM, such as staffing interconnect university, industry, oping researchers’ capabilities to regional development and solution and government groups co-investing address the grand challenge problems architecture labs, sales and consult- in creating capabilities that can dra- of business and society. ing centers, service delivery centers, matically improve the productivity and business analytics and optimi- of future knowledge workers. These Readiness zation centers, as well as diverse trends are indicative of the central The Academic Initiative program industry and technology specialist and growing role that academic lets registered faculty access IBM positions.

AUGUST 2010 103

INDUSTRY PERSPECTIVE

Revenue ulty, members of advisory boards, or broad communications skills will From IBM’s business perspective, university ambassadors, sharing their be needed (http://beyond-it-inc.com/ universities are themselves complex knowledge and skills to enrich uni- GKEblog/ibm-transforms-academic-______organizations, often with tens of versity courses and research projects. initiative-to-ssme-to-smarter-planet.______thousands of “customers” on campus, Dozens of other programs, includ- html).____ Working with universities, and many more accessing online con- ing the World Community Grid, IBM has identified a $4 trillion chal- tent or global campus locations. IT benefit from open collaborative lenge, requiring a system-of-systems continues evolving to better address research efforts in universities and approach (www.ibm.com/services/ these needs. other community organizations us/gbs/bus/html/ibv-smarter-planet-______Across content management, around the world. Further, many IBM system-of-systems.html).______Universities asset tracking, campus data center executives participate in the Partner- as centers of innovation, sources of design, operations, enterprise and ship Executive Program (PEP) for talent, and cities within cities are ide- architecture planning, and organiza- universities, which identifies collabo- ally positioned to play a significant tional change execution, many IBM ration opportunities. role in addressing the $4 trillion offerings attempt to help universities Annually, these programs result in dollar challenge to accelerate quality- more efficiently create new capabili- tens of millions of dollars and hun- of-life improvements in major urban ties for their students, faculty, and dreds of thousands of hours invested areas around the world. This is a administrators. in building multifaceted relationships function of both the quality of ser- For example, as universities with more than 5,000 universities vice from systems and the quality of work to reengineer and expand and institutions of higher education opportunities for people in systems, capabilities, more public and pri- around the world. Equally signifi- including meaningful jobs and gainful vate partnership opportunities cant, these collaborations provide investments. present themselves, including co-investment opportunities to uni- Fundamental theoretical con- energy-efficient buildings and data versities and other partners that can tributions to mathematics, physics, centers, university-based service significantly advance knowledge in computer science, and the emerging delivery centers, high-performance emerging areas. transdiscipline of service science, computing research centers, regional management, and engineering (SSME) nanotechnology centers, business provide additional vibrant areas of analytics centers, and a host of other uring Earth Day 2009, interest. While the bulk of the IBM projects that enhance universities’ IBM invited universities university programs’ efforts focuses prestige and create more regional, D from around the world to on applied research areas, basic high-skill jobs. participate in IBM’s Smarter Planet research remains essential for open- University Jam, an online brainstorm- ing important new vistas. Together, Responsibility ing session. For three days, nearly we can continue to accelerate innova- IBM, like other leading global cor- 2,000 students and faculty interacted tion that matters throughout business porations, encourages its employees with IBM staff members, generating and society, improving the quality of to give back to their local communi- discussion topics and ideas about life around the world. ties through charitable donations, smarter healthcare, education, energy volunteerism, and other corporate and electricity, water and the environ- James C. Spohrer is director of IBM social-responsibility programs, ment, and city planning. Students’ University Programs (IBM UP) world- including IBM’s own Corporate Ser- readiness to engage with industry and wide, based at the IBM Almaden vice Corps. A top corporate priority is government to begin making a differ- Research Center in San Jose, Califor- to ensure the strength and diversity of ence in reducing waste, eliminating nia. Contact him at [email protected].______the STEM (science-technology-engi- inefficiencies, and realizing the ben- ___com. neering-mathematics) pipeline into efits of instrumented, interconnected, universities, with programs reaching and intelligent systems proved to be a Editor: Sumi Helal, Department of Computer from primary through secondary key finding (http://download.boulder. and Information Science and Engineering, education and into lifelong profes- ibm.com/ibmdl/pub/software/dw/ University of Florida; [email protected] sional learning. university/smartplanet/Jam_Report______IBM employees give generously 2009.pdf).______as alumni, and corporate matching In the decade ahead, IBM will seek programs increase these benefits. university partners ready to build a Selected CS articles and columns Thousands of them volunteer their smarter planet. T-shaped graduates are available for free at ____http:// time as guest lecturers, adjunct fac- with deep problem-solving skills and ComputingNow.computer.org.

104 COMPUTER

Limited Time Offer Half Year Subscriptions Available Now!

See how the Computer Society publications are leading the way to new discoveries.

There’s still time to subscribe to your favorite Computer Society magazines and journals. You can also select books, ReadyNotes and EssentialSets on your speciﬁc topics of interest.

Go to: www.computer.org/promos/HY10OMNI Orders must be received by 10 August 2010.

INVISIBLE COMPUTING The Hardware Is Not a Given

Steve Hodges and Nicolas Villar, Microsoft Research Cambridge

Commodity hardware enables a range of cost-effective ubiquitous computing research, but innovative custom hardware can drive compelling new user experiences.

ne of the dreams of ferent forms that support rich and photography, and, most prevalent of ubiquitous computing varied means of interaction. all, an array of mobile communica- is an ecology of het- In many ways, we’re still a long way tion devices and services. As these O erogeneous computer from achieving ubiquitous computing new application areas become estab- systems, built into the environment (A. Schmidt, “Ubiquitous Comput- lished, the limitations of traditional around us and acting in concert ing: Are We There Yet?” Computer, computer hardware are apparent. It’s to make our lives easier and more Feb. 2010, pp. 95-97). Nonetheless, clear that a diverse set of technolo- fun. Fundamental to this vision is during the past few years, comput- gies, form factors, and interaction a movement away from the desk- ing devices have come to increasingly techniques is becoming appropriate. top computing metaphor, in which leverage characteristics associated The community of researchers and a single user interacts very inten- with ubiquitous computing such as developers who are exploring new tionally via a display, keyboard, and mobility, real-time connectivity, and directions for future generations of mouse. Instead, many researchers sensing. Examples include satellite nondesktop computing systems has imagine computers with very dif- navigation, computer gaming, digital a difficult challenge—to build compelling new applications that make lives easier and more enjoyable as well as consider new technologies, form factors, and interaction techniques. Innovative software running on traditional computer hardware isn’t always enough; there’s a lot of value in pushing the boundaries all the way up and down the stack: from hardware, through software, and ultimately to the user experience.

TWO EXAMPLES Movement-based game consoles such as the Nintendo Wii, with its motion-sensitive controller, and the recently announced Xbox 360 Kinect, which tracks body motion without a controller at all, are good examples of how changes in the hardware Figure 1. The motion-based controller has created new markets for computer game and form factor can be coupled with consoles with the old and young alike. innovative software to deliver a revo-

lutionary user experience. In both cases, the designers combined fundamentally different electronic sensing technologies, such as motion sensors and vision systems, with novel software techniques ranging from signal processing to computer graphics. These technologies are creating new markets for computer game consoles, as Figure 1 shows. Another example of how relatively small hardware changes can affect an entire market is evident in cellular telephony. Just a few years ago, there Figure 2. The SenseCam wearable camera has a wide-angle lens and takes photos were essentially two established automatically, creating a visual diary of the wearer’s day. form factors for mobile phones: the clam shell and the candy bar. How- ever, advances in multitouch sensing to create something revolutionary. type is essential to its usability. For enabled this technology to migrate Developing custom hardware requires memory-impaired patients who have to mobile phones. The richer inter- considerable expertise, time, and found the device invaluable, a smart- action afforded by multitouch input money, but the flexibility it affords phone-based implementation simply largely removed the need for physical makes it worthwhile. Put simply, there wouldn’t deliver an acceptable user keys, allowing the display to cover the are a great many things that you can’t experience. device’s entire front surface—thus a do with a smartphone. new form factor was born. With the WISP inclusion of simple sensing technolo- SenseCam There are many other examples of gies such as an accelerometer, the SenseCam, shown in Figure 2, the importance of bespoke hardware smartphone enables a range of com- is a small, wearable camera with a under the broad umbrella of ubiqui- pelling new user experiences. wide-angle lens that takes digital still tous computing. Intel Labs’ Wireless images automatically (S. Hodges et al., Identification and Sensing Platform BEYOND SMARTPHONES “SenseCam: A Retrospective Memory (J.R. Smith et al., “A Wirelessly- The research community is natu- Aid,” UbiComp 2006: Ubiquitous Com- Powered Platform for Sensing and rally drawn to commodity hardware puting, LNCS 4206, Springer, 2006, pp. Computation,” UbiComp 2006: Ubiqui- such as the smartphone and motion- 177-193). The camera doesn’t have a tous Computing, LNCS 4206, Springer, sensitive game controller. These viewfinder or display—it simply cap- 2006, pp. 495-506) is essentially devices have enabled many new tures photos of whatever the wearer a special type of radio-frequency types of applications without any is looking at whenever triggered by identification tag that contains a requirement to develop new hard- its internal timer, motion, light-level, general-purpose microcontroller in ware. Not only do they offer great or temperature sensors. SenseCam is place of the inflexible hard-coded value, but in many cases they readily designed for very-low-power opera- integrated circuit used in commer- support interoperability that can be tion so it can run continuously all cial RFID tags. The WISP’s ability to leveraged to scale up research efforts. day, resulting in a visual record that harvest power from a nearby RFID Smartphones in particular provide a is created without any intervention or reader enables a new class of bat- standardized platform on which to conscious attention from the wearer. tery-less sensing applications very develop, making it easier to run large- In many ways the SenseCam hard- different from anything that could scale deployments of probes and ware is very straightforward. Indeed, be performed with commodity hard- prototypes. This in turn generates at first blush it would appear that a ware, let alone a smartphone. deeper insights than would have been smartphone could readily be repur- possible with custom-built hardware. posed as a SenseCam. However, the Peppermill But commodity hardware isn’t a particular combination of ease of Another example of power- silver bullet. If we simply build new use, wearability, robustness, thin harvesting research enabled through applications on existing hardware, we form factor, wide-angle field of view, new hardware is the recent Pepper- will often limit ourselves to evolution- and lengthy battery lifetime offered mill project (N. Villar and S. Hodges, ary progress, missing the opportunity by our custom-built research proto- “The Peppermill: An Interaction-

AUGUST 2010 107

INVISIBLE COMPUTING

sary to develop new hardware from scratch for every project. Sometimes a useful approach is simply to imagine what would be possible if different sensing systems and hardware form factors were available and reinforce this thought process by building Wizard-of-Oz prototypes. For example, researchers can use infrastructure-heavy indoor location-sensing systems to explore applications that respond to the vary- ing absolute and relative positions of multiple users and mobile devices. The Active Bat system (M. Addle- see et al., “Implementing a Sentient Figure 3. The Peppermill is an interaction-powered user interface device that can be Computing System,” Computer, Aug. used to control a TV. Rotating the Peppermill clockwise with the red button pressed 2001, pp. 50-56) enabled numerous turns up the volume, and with the green button pressed increments the channel. location-based applications like that shown in Figure 4 to be prototyped and deployed in a lab environment. Similarly, optical motion-capture systems such as that produced by Vicon (http://vicon.com) let research labs create an environment for evaluat- ing new applications and interaction techniques. Another alternative to developing custom hardware is to use someone else’s. Microsoft Research loaned out several hundred SenseCam prototypes to life-sciences clinicians and researchers, and more than 40 labs worldwide now use the device to study its effects on normal memory and on people with memory impair- ment. The images turn out to be a powerful memory aid; review- ing photos captured by SenseCam frequently elicits powerful recall, Figure 4. The Active Bat system enabled the prototyping and deployment of various rekindling the thoughts and feelings location-based applications. the wearer experienced at the time. The results have been so successful Powered User Interface Device,” Proc. volume and channel, simultaneously that the device is now available com- 4th Int’l Conf. Tangible, Embedded, and generating all the energy required to mercially to support global research Embodied Interaction, ACM Press, power the device. (http://viconrevue.com). 2010; http://tei-conf.org/10/uploads/ Intel has likewise made the WISP Program/p29.pdf).______The simple ALTERNATIVES TO platform available to numerous and ubiquitous TV remote control CUSTOM HARDWARE researchers to explore various appli- inspired the project, which couples Ubiquitous computing research- cations and new forms of interaction a novel interaction metaphor with a ers face a dilemma: exploring new that aren’t possible using standard new device form factor. As Figure 3 hardware designs and configurations hardware. And Microsoft Research shows, the user “grinds” the pepper- is clearly valuable, but it can also be has made the power harvesting mill-shaped device to select the TV expensive. However, it isn’t neces- and direction-and-speed-detection

108 COMPUTER

circuitry used in the Peppermill available as a separate Peppermill power board, shown in Figure 5. The complete design informa-

tion is now available online (http://______research.microsoft.com/peppermill) for anyone to experiment with and adapt to their own application ideas. Easy-to-use and well-documented platforms like Arduino (http://arduino.

cc)__ are making it increasingly straightforward for researchers and developers with little electronics development experience to prototype custom devices. The required building blocks, which include a variety of modules, sensors, actuators, and discrete electronic Figure 5. Peppermill power board. Microsoft Research has distributed more than 100 of components are available ex-stock these to the research and hobbyist communities to encourage exploration. from outlets such as SparkFun (http://sparkfun.com). Websites like MAKE (http://make-______ith so many resources Nicolas Villar is a researcher in the zine.com)______and Instructables (http://____ available to the research Sensors and Devices Group at Micro- instructables.com) have created W and development commu- soft Research Cambridge, focusing growing communities that generate nity, there are growing opportunities on the development of new hardware platforms and tools to enable valuable resources for ubiquitous to advance toward the ubiquitous technical innovation. Contact him at computing researchers in the form computing vision. It isn’t clear exactly [email protected].______of detailed instructions, code sam- where ubiquitous computing is going ples, and sample schematics for an to take us, but one thing is certain: the incredibly diverse set of creative hardware is not a given. projects. In many cases, these mate- Editor: Albrecht Schmidt, Institute for Computer Science and Business Information rials can be readily adapted and Steve Hodges manages the Sen- Systems, University of Duisburg-Essen, extended to enable research beyond sors and Devices Group at Microsoft what is possible using commodity Research Cambridge, UK. Contact him Germany; [email protected]______hardware. at [email protected].______

Silver Bullet Security Podcast

In-depth interviews with security gurus. Hosted by Gary McGraw.

www.computer.org/security/podcasts______