BERKELEY ODYSSEY

Ten years of BSD history

by Marshall Kirk McKusick

Ken Thompson and Dennis Ritchie presented the first UNIX paper at the Symposium on Operating Systems Principles at Pur­ due University in November, 1973. Professor Bob Fabry was in attendance and immediately became interested in obtaining a f copy of the system to experiment with at Berkeley. - At the time, Berkeley had only large mainframe computer systems doing batch processing, so the first order of business was to get a PDP-11/45 suitable for running the then current Version 4 of UNIX. The Computer Science Department, together with the Mathematics Department and the Statistics Department were able to jointly purchase a PDP-11/45. In January, 1974, a Version 4 tape was delivered and UNIX was installed by graduate student ¥<~ Keith Standiford. Although Ken Thompson was not involved in the installation — as he had been for most systems up to that time — his exper­ tise was soon needed to determine the cause of several strange system crashes. Because Berkeley had only a 300 baud acoustic- coupled modem without auto answer capability, Thompson would call Standiford in the machine room and have him insert the phone into the modem; in this way Thompson was able to remote­ ly debug crash dumps from New Jersey.

9mm £££

SK wmmMm&?<%?+•*>' '-•fy&Z j?m gis£^t - WMm ^ssg®mmm

m^f, ?*S£ SP mm? r^^^m^^S?^^ :%%&m.

':•! '-^SS&^i

* Jfjfe BERKELEY ODYSSEY

Many of the crashes were project became the first group in immediate interest in the new caused by the disk controller's the Computer Science department system. Initially they began work­ inability to reliably do overlapped to distribute its software. Several ing on a Pascal system that seeks, contrary to the documenta­ hundred Ingres tapes were Thompson had hacked together tion. Berkeley's 11/45 was among shipped over the next six years, while hanging around the 11/70 the first systems that Thompson helping to establish Berkeley's machine room. They expanded had encountered that had two reputation for designing and and improved the interpreter disks on the same controller! building real systems. system to the point that it became Thompson's remote debugging Even with the departure of the programming system of was the first example of the the Ingres project from the 11/45, choice for students because of its cooperation that sprang up be­ there was still insufficient time excellent error recovery scheme tween Berkeley and Bell Labs. The and fast compile and execute willingness of the researchers at time. the Labs to share their work with With the replacement of Berkeley was instrumental in the Model 33 teletypes by ADM-3 r rapid improvement of the softw are Arriving in the Fall screen terminals, Joy and Haley available at Berkeley. began to feel stymied by the con­ Though UNIX was soon of 1975 were two straints of the ed editor. Working reliably up and running, the coa­ from an editor named em that lition of Computer Science, unnoticed graduate they had obtained from Professor Mathematics, and Statistics began George Coulouris at Queen Mary's to run into problems: Math and students, Bill Joy College in London, they worked to Statistics wanted to run DEC's produce the line-at-a-time editor RSTS system. After much debate, and Chuck Haley. ex. a compromise was reached in With Ken Thompson's depar­ which each department would get ture at the end of the Summer of an eight-hour shift: UNIX would 1976, Joy and Haley begin to take run for eight hours followed by 16 an interest in exploring the inter­ hours of RSTS. To promote available for the remaining stu­ nals of the UNIX kernel. Under fairness, the time slices were dents. To alleviate the shortage, Schriebman's watchful eye, they rotated each day. Thus UNIX ran Professors Michael Stonebraker first installed the fixes and im­ 8 am to 4 pm one day, 4 pm to and Bob Fabry set out in June, provements provided on the "fif­ midnight the next day, and mid­ 1974. to get two instructional ty changes" tape from Bell Labs. night to 8 am the third day. 1 l/45s for the Computer Science Having learned to maneuver Despite the bizarre schedule, department's own use. Early in through the source code, they sug­ students taking the Operating 1975. the money was obtained. At gested several small enhance­ Systems course preferred to do nearly the same time DEC an­ ments to streamline certain kernel their projects on UNIX rather than nounced the 11/70, a machine bottlenecks. on the batch machine. that appeared to be much superior Professors Eugene Wong and to the 11/45. Money for the two FIRST DISTRIBUTIOM Michael Stonebraker were both 1 l/45s was pooled to buy a single Meanwhile, interest in the er­ stymied by the confinements of 11/70 that arrived in the Fall of ror recovery work in the Pascal the batch environment, so their 1975. Coincident with the arrival compiler brought in requests for Ingres database project was of the 11/70, Ken Thompson copies of the system. Early in among the first groups to move decided to take a one-year sab­ 1977, Joy put together the from the batch machines to the in­ batical as a visiting professor at "Berkeley Software Distribution". teractive environment provided his alma mater. Thompson, This first distribution included the by UNIX. They quickly found the together with Jeff Schriebman Pascal system and — in an shortage of machine time and the and Bob Kridle, brought up the obscure subdirectory of the Pascal odd hours on the 11/45 in­ latest UNIX, Version 6, on the source — the editor ex. Over the tolerable, so in the Spring of 1974, newly installed 11/70. next year, Joy, acting in J.he they purchased an 11/40 running Also arriving in the Fall of capacity of distributiojijsecretary, the newly available Version 5. 1975, were two unnoticed sent out about 30 free copies of the With their first distribution of In­ graduate students. Bill Joy and system. gres in the Fall of 1974, the Ingres Chuck Haley; they both took an With the arrival of some 32 UNIX REVIEW JANUARY 1985 BERKELEY ODYSSEY

ADM-3a terminals offering screen- Fateman and 13 other faculty having his group write a virtual addressable cursors, Joy was members put together an NSF memory system for UNIX. Ozalp finally able to write vi, bringing proposal that they combined with Babaoglu, one of Ferrari's stu­ screen-based editing to Berkeley. some departmental funds to pur­ dents, set about to find some way He soon found himself in a quan­ chase a VAX. of implementing a working set dary. As is frequently the case in Initially the VAX ran DEC's paging system on the VAX; his universities strapped for money, operating system VMS, but the task was complicated because the old equipment is never replaced department had gotten used to the VAX lacked reference bits. all at once. Rather than support UNIX environment and wanted to As Babaoglu neared the com­ code for optimizing the updating continue using it. So, shortly after pletion of his first cut at an im­ of several different terminals, he plementation, he approached Bill decided to consolidate the screen Joy for some help in understand­ management by using a small ing the intricacies of the UNIX interpreter to redraw the screen. Initially the VAX kernel. Intrigued by Babaoglu's This interpreter was driven by a approach, Joy joined in helping to description of the terminal char­ ran DEC'S integrate the code into 32/V and acteristics, thus spawning the then with the ensuing debugging. now famous termcap. operating system Unfortunately, Berkeley had By mid-1978, the software only a single VAX for both system distribution clearly needed to be VMS, but the development and general produc­ updated. The Pascal system had tion use. Thus for several weeks, been made markedly more robjast department had the tolerant user community alter­ through feedback from its expand­ nately found themselves logging ing us^_conimjunity, and had gotten used to the into 32/V and "Virtual VAX/ been split intojtwo passes so that UNIX". Often their work on the IF could be run on PDP"-ll/34s. UNIX environment latter system would come to an The result of the update was the abrupt halt, followed several "Second Berkeley Software and wanted to minutes later by a 32/V login Distribution" that was quickly prompt. By January of 1979, most shortened to 2 BSD. Along with continue using it. of the bugs had been worked out, the enhanced Pascal system, vi and 32/V had been relegated to and termcap for several terminals history. was included. Once again, Bill Joy Joy saw that the 32-bit VAX single-handedly put together the arrival of the VAX, Fateman would soon obsolete the 16-bit distributions, answered the obtained a copy of the 32/V port of PDP-11 and began to port the 2 phone, and incorporated user UNIX to the VAX by John Reiser BSD software to the VAX. While feedback into the system. Over the and Tom London of Bell Labs. Peter Kessler and I ported the next year, nearly 75 tapes were Although 32/V provided a Pascal system, Joy ported the shipped. Though Joy moved on to Version 7 UNIX environment on editors ex and vi, the C shell, and other projects the following year, the VAX, it did not take advantage the myriad other smaller pro­ the 2 BSD distribution continued of the virtual memory capability of grams on 2 BSD. By the end of to expand. Today the latest ver­ the VAX hardware. Like its 1979, a complete distribution had sion of this distribution, 2.9 BSD, predecessors on the PDP-11, it been put together. This distribu­ is a complete system for PDP-1 Is. was entirely a swap-based system. tion included the virtual memory For the Macsyma group at kernel, the standard 32/V utilities, VAX UNIX Berkeley, the lack of virtual and the additions from 2 BSD. In Early in 1978, Professor memory meant that the process December, 1979, Joy shipped the Richard Fateman began looking address space was limited by the first of nearly a hundred copies of for a machine with a larger ad­ size of the physical memory, 3 BSD, the first VAX distribution dress space that he could use to initially 1 MB on the new VAX. from Berkeley. continue his work on Macsyma To alleviate this problem, that had started on a PDP-10. The Fateman approached Professor DARPA SUPPORT newly announced VAX-11/780 Domenico Ferrari, a member of Meanwhile, in the offices of seemed to fulfill the requirements the systems faculty at Berkeley, the planners for the Defense Ad­ and was available within budget. to investigate the possibility of vanced Research Projects Agency, 34 UNIX REVIEW JANUARY 1985 if.ftti BERKELEY ODYSSEY

DARPA, discussions were being that time, DARPA was finding that machines. In addition, many sites held that would have a major in­ many of the computers at these were unable to share their soft­ fluence on the work at Berkeley. centers were reaching the end of ware because of the diversity of One of DARPA's early successes their useful lifetime and had to be hardware and operating systems. had been to set up a nationwide replaced. The heaviest cost of Choosing a single hardware computer network to link together replacement was the porting of the vendor was impractical because of all its major research centers. At research software to the new the widely varying computing needs of the research groups and the undesirability of depending on a single manufacturer. Thus, the planners at DARPA decided that the best solution was to unify at the operating systems level. After much discussion, UNIX was chosen as a standard because of its proven portability. In the Fall of 1979, Bob Fabry responded to DARPA's interest in moving towards UNIX by writing a proposal suggesting that Berkeley develop an enhanced version of 3 BSD for the use of the DARPA community. Fabry took a copy of his proposal to a meeting of DARPA image processing and VLSI contractors, plus represen­ tatives from Bolt, Beranek, and Newman, the developers of the ARPAnet. There was some reser­ vation whether Berkeley could produce a working system, but the release of 3 BSD in Decem­ ber, 1979, assuaged most of the Without SoftSheli; doubts. With the increasingly good learning to live with UNIX™ reputation of the 3 BSD release to could be a real trap. validate his claims, Bob Fabry was able to land an 18-month contract SoftShell is a convenient interface which guides you gracefully with DARPA beginning in April, through UNIX. If you're new to UNIX, or have the task of training 1980. This contract was to add new users, SoftShell simplifies its complexity. If you're already a features needed by the DARPA UNIX fan, you'll love SoftShell because it augments your system. contractors. He immediately hired You'll find yourself further exploring the great depth and ver­ Laura Tong to handle the project satility of UNIX. Available for UNIX System-Y, Berkeley 42 and administration. With the negotia­ IBM's PC/IX, SoftShell is an invaluable tool for users at all levels tions for the contract on track, of expertise. For information, contact Logical Software, Inc., Fabry turned his attention to find­ 17 Mount Auburn Street, Cambridge, MA 02138, (617) 864-0137. ing a project leader to manage the UNIX is a trademark of AT&T Bell Laboratories. software development. Fabry had L assumed that since Joy had just passed his Ph.D. qualifying ex­ ifjf^ Logical Software Inc. amination, he would rather con­ LLLL centrate on completing his degree than assume the software develop­ ment position. But Joy had other Circle No. 16 on Inquiry Card See us at UniForum Booth # 1279 plans. One night in early March he 36 UNIX REVIEW JANUARY 1985 BERKELEY ODYSSEY

phoned Fabry at home to express lifetime about 400 distributions system could easily participate in interest in taking charge of the fur- were shipped. the ARPAnet. ther development of UNIX. To assist in defining the new Though surprised by the offer, 4.2 BSD system, Duane Adams, Berkeley's Fabry took little time to agree. With the release of 4.1 BSD, contract monitor at DARPA, The project started promptly. much of the furor over perfor­ formed a group known as the Tong set up a distribution system mance died down. DARPA was "steering committee" to help that could handle a higher volume sufficiently satisfied with the guide the design work and ensure of orders than Joy's previous results of the first contract that a that the research community's distributions. Fabry managed to new two-year contract was needs were addressed. This com­ coordinate with Bob Guffy at granted to Berkeley with funding mittee met twice a year between AT&T, and lawyers at the Uni­ April, 1981 and June, 1983, and versity of California to formally included Bob Fabry, Bill Joy, release UNIX under terms agree­ and Sam Leffler of the University able to all. Joy incorporated Jim of California at Berkeley; Alan Kulp's job control, added auto With the release of Nemeth and Rob Gurwitz of Bolt, reboot, a IK block file system, and Beranek, and Newman; Dennis support for the latest VAX 4.1 BSD, much of Ritchie of Bell Laboratories; machine, the VAX-11/750. By Keith Lantz of Stanford Uni­ October, 1980, a polished distribu­ the furor over versity; Rick Rashid of Carnegie- tion that also included the Pascal Mellon University; Bert Halstead compiler, the Franz Lisp system, of Massachusetts Institute of and an enhanced mail handling performance died Technology; Dan Lynch of The system was released as 4 BSD. Information Sciences Institute; J3uring_its; nine^nonth lifetime, down. Duane Adams and Bob Baker of nearly 150 copies were shipped. DARPA; and Jerry Popek of the The license arrangement wasjon University of California at Los ajperinstitution basis rather than Angeles. Beginning in 1984, these a per machine basis, thus the almost five times that of the meetings were supplanted by distribution ran on about 500 original. Half of the money went to workshops that were expanded to machines. the UNIX project, the rest to other include many more people. With the increasingly wide researchers in the Computer An initial document pro­ distribution and visibility of Science department. The contract posing facilities to be included in Berkeley UNIX, several critics called for major work to be done the new system was circulated to began to emerge. David Kashtan on the system so the DARPA the steering committee and other at Stanford Research Institute research community could better people outside Berkeley in July, wrote a paper describing the do its work. 1981, sparking many lengthy results of benchmarks he had run Based on the needs of the debates. In the Summer of 1981, on both VMS and Berkeley UNIX. DARPA community, goals were I became involved with the project These benchmarks showed set and work began to define the and took on the implementation of several severe performance prob­ modifications to the system. In the new file system. During the lems with the UNIX system for the particular, the new system was summer, Joy concentrated on im­ VAX. Setting his future plans expected to include a faster plementing a prototype version of aside for several months, Joy file system that would raise the interprocess communication systematically began tuning up throughput to the speed of facilities. In the Fall of 1981, Sam the kernel. Within weeks he had available disk technology, would Leffler joined the project as a full- a rebuttal paper written showing support processes with multi- time staff member to work with that Kastan's benchmarks could gigabyte address space require­ Bill Joy. be made to run as well on UNIX as ments, would provide flexible When Rob Gurwitz released they could on VMS. Rather than interprocess communication an early implementation of the continue shipping 4 BSD, the facilities that would allow re­ TCP/IP protocols to Berkeley, Joy tuned up system with the addition searchers to do work in integrated it into the system and of Robert Elz's auto configuration distributed systems, and would tuned its performance. During this code was released as 4.1 BSD in integrate networking support so work, it became clear to Joy and June, 1981. Over its two-year that machines running the new Leffler that the new system would 38 UNIX REVIEW JANUARY 1985 ft.tXki BERKELE Y ODYSSEY

need to provide support for more file systems to be dumped and of the project to Professors than just the DARPA standard restored to convert from 4.1a to Domenico Ferrari and Susan network protocols. Thus, they 4. lb. Once the file system proved Graham to begin a sabbatical free redesigned the internal structur­ to be stable, Leffler proceeded to from the frantic pace of the ing of the software, refining the add the new file system-related previous four years. Leffler con­ interfaces so that multiple net­ system calls, while Joy worked on tinued the completion of the work protocols could be used revising the interprocess com­ system, implementing the new simultaneously. munication facilities. signal facilities, adding to the With the internal restructur­ In late Spring 1982, Joy networking support, redoing the ing completed and the TCP/IP announced he was joining Sun standalone I/O system to simplify protocols integrated with the Microsystems. Over the summer, the installation process, in­ prototype IPC facilities, several he split his time between Sun and tegrating the disk quota facilities simple applications were created Berkeley, spending most of his from Robert Elz, updating all the to provide local users access time polishing his revisions to the documentation, and tracking the to remote resources. These pro­ interprocess communication bugs from the 4.1c release. In grams, rep, rsh, rlogin, and rwho, facilities and reorganizing the August, 1983, the system was were intended to be temporary UNIX kernel sources to isolate released as 4.2 BSD. tools that would eventually be machine dependencies. Pauline When Leffler left Berkeley for replaced by more reasonable Schwartz was hired to take over Lucasfilm following the comple­ facilities (hence the use of the the distribution duties. David tion of 4.2, he was replaced by distinguishing r prefix). This Mosher was hired as a technical Mike Karels. Karels's previous system, called 4.1a, was first manager to resolve problems from experience with the 2.9 BSD soft­ distributed in April, 1982 for local users in the field and to handle ware distribution provided an use; it was never intended that it ordering, installation, and run­ ideal background for his new job. would have wide circulation, ning of the project's hardware. The popularity of 4.2 BSD was im­ though bootleg copies of the With Joy's departure, Leffler pressive; within 18 months, more system proliferated as sites grew took over responsibility for com­ copies of 4.2 BSD had been ship­ impatient waiting for the 4.2 pleting the project. Certain ped than of all the previous release. deadlines had already been Berkeley software distributions The 4. la system was obsolete established and the release had combined. long before it was frozen. been promised to the DARPA As with 4 BSD, commentary However, its construction and community for the Spring of 1983. of the vociferous critics was quick feedback from users provided Given the time constraints, the in coming. Most of the complaints valuable information that was us­ work remaining to complete the indicated that the system ran too ed to create a revised proposal for release was evaluated and slowly. The problem, not surpris­ the new system called the "4.2 priorities were set. In particular, ingly, was that the new facilities BSD System Manual". This docu­ the virtual memory enhance­ had not been tuned and that many ment was circulated in February, ments and the most sophisticated of the kernel data structures were 1982 and contained a concise parts of the interprocess com­ not well suited to their new uses. description of the proposed user munication design were relegated Karels' first year on the project interface to the system facilities to low priority (and later shelved was spent tuning and polishing that were to be part of 4.2 BSD. completely). Also, with the im­ the system. An anticipated release Concurrent with the 4.1a plementation more than a year old of the polished system early in development, I completed the and the UNIX community's ex­ 1985 is expected to quell many of implementation of the new file pectations heightened, it was the performance complaints — system, and by June of 1982 had decided an intermediate release much as the 4.1 BSD release quell­ fully integrated it into the 4.1a should be put together to hold peo­ ed many of the complaints about kernel. The resulting system was ple until the final system could be 4 BSD. called 4. lb and ran on only a few completed. This system, called After completing my Ph.D. in select development machines at 4.1c, was distributed in April, December 1984, I joined Mike Berkeley. Joy felt that with signifi­ 1983; many vendors used this Karels on the project. We hope cant impending changes to the release to prepare for ports of 4.2 that other researchers will con­ system, it was best to avoid even to their hardware. tinue to share their work with a local distribution, particularly In June, 1983, Bob Fabry Berkeley. By incorporating the since it required every machine's turned over administrative control work of other researchers 40 UNIX REVIEW JANUARY 1985 BERKELEY UNDERGROUND

uniformly into the UNIX system at from describing the enormous tronic System Command under Berkeley, we can continue to offer amount of work that went into the Contract No. N00039-82-C-0235. the UNIX community a widely user level software. I thank those available state-of-the-art UNIX people whose accomplishments system. have not been mentioned, but Kirk McKusick is involved in nevertheless have contributed the development of Berkeley UNIX ACKNOWLEDGEMENTS much to the system's vitality. as a Research Computer Scientist for / thank Bill Joy, Sam Leffler, For the last five years, I have the Computer Systems Research Eric Allman, and Professors Bob been supported by the National Group at the University of Califor­ Fabry, Richard Fateman, and Science Foundation under grant nia. While a graduate student, he im­ Domenico Ferrari for providing MCS80-05144, and the Defense plemented the fast file system the historical information pre­ Advance Research Projects Agen­ distributed on 4.2 BSD and worked sented in this article. Unfortunate­ cy (DoD) under ARPA Order No. on the Berkeley Pascal system. • ly, space constraints prevent me 4031 monitored by Naval Elec­ FEAR AND LOATHING ON THE "f UNIX TRAIL *76

It was 2 am and I was lying mal order. We might have called face down on the floor in Cory Notes from our group "Berkeley Undergrade \ Hall, the EECS building on the UC the underground uate Programmers Sor a Better^ Berkeley campus, waiting for Bob UNIX", or* less euphemistically^ to finish installing our bootleg *TrustratedH2Ktoreffrt)urO :l copy of the UNIX kernel. If suc­ by Doug Merrltt Ideas". But to truth, our group cessful, new and improved ter­ with Ken Arnold and Bob Toxen was never named. It was simply a minal drivers we had written matter of Us versus Them. would soon be up and running. 'Them** was the bureaucracy We were enhancing the — the school administrator, the system in the middle of the night system administrators, most pro- " because we had no official sanc­ fessors, some grad students, and ^ tion to do the work. That didn't even the legendary Implementors stop us, though, since UNIX had themselves at Bell Labs. just freshly arrived from Bell Labs, ••Us" was* small, ielf- where computer security had selected group of undergraduates never been an issue. The system with a passion for UNIX. We were 'i was now facing its first acid test — interested in computers and In exposure to a group of intelligent, programming because it fasci­ determined students — and its nated us; we Bved for fee high security provisions were failing level of intellectual, stimulation 1 with regularity. only haciln^^couW pr^vfctel | I was lying face down because Although some In our group never; I'd gone without sleep for over two expressed an Interest'in breaking days, and the prone position computer TOcurttyi^others^Ifr, somehow seemed the most logical vested thousands of fruitful hours under the circumstances. Bob was to stealing accounts and gaining still working because he'd napped superuser access tb^&iious UNIX not 30 hours before, giving him systems. 0ur^object? -To"read A seniority under the "Hacker-best- source able-to-perform" rule of our infor- is

42 UNIX REVIEW JANUARY 1985 Circle No. 20 on Inquiry Card • BERKELEY UNDERGROUND

out of trouble, although one of our We modified the kernel to wherein a group of people will rank once had his phone records support asynchronous I/O, distri­ tend to ignore ideas originated subpoenaed by the FBI — after a buted files, security traces, "real­ outside their own social group. minor incident with a Lawrence time" interrupts for subprocess However, there was a stronger Livermore National Laboratory multitasking, limited screen force at work at Berkeley, where computer. The Feds seemed to editing, and various new system a certain social stratification think our comrade had been did­ prevails that finds Nobel Lau­ dling with top secret weapons reates and department chairs research, but he actually hadn't. ranking as demigods, professors Our group could probably functioning as high priests, best be characterized by its in­ It was simply a graduate students considered as terest in creating and using power­ lower class citizens, and under­ ful software, regardless of the matter of Us versus graduates existing only on suf­ source of the idea. Our battle ferance from the higher orders — cry, thanks to Ross Harvey, was Them. and suffered very little at that. "FEATURES!!!", and we took it Now, the individuals cannot be seriously. Well, Ross may have blamed for what is, in essence, an been a little sarcastic about it, entire social order. But this is not since he was referring to super­ calls. We wrote compilers, ass­ to say that we did not hold it fluous bells and whistles. But I emblers, linkers, disassemblers, against them — for we most used the expression as shorthand database utilities, cryptographic assuredly did. Unfortunately, it for "elegant, powerful, and flexi­ utilities, tutorial help systems, took time for us to appreciate the ble". We were always bugging games, and screen-oriented ver­ difficulties of Fighting City Hall. Them to add "just one more sions of standard utilities. User This is why We were feature" to some utility like the friendly utilities for new users that frustrated. This is why We felt We shell or kernel. Although They ac­ avoided accidental file deletion, HAD to break security. Once We cepted some suggestions. They libraries to support common did, We simply added Our features didn't think twice about most. operations on data structures to the system, whether The One example stands out. In such as lists, strings, trees, sym­ Powers That Be liked it or not. early 1977, Ross, Bob, and I spent bol tables, and libraries to perform Needless to say, They didn't. This months collaborating on a new arbitrary precision arithmetic and is why We felt like freedom and improved shell, just before symbolic mathematics were other fighters, noble figures even when Bill Joy had started on what is contributions. We suggested im­ found in the ignoble position of ly­ now known as the C shell. The provements to many system calls ing face down on the floor of Cory most historically significant and to most utilities. We offered to Hall at two in the morning. features we designed were Ross's fix the option flags so that the dif­ We were on a mission that command to change the shell's ferent utilities were consistent morning to install our new ter­ prompt, Bob's command to print with one another. minal driver. With the old, stan­ or chdir to the user's home direc­ To Us, nothing was sacred, dard terminal driver, the screen tory, and my own edit feature, and We saw a great deal in UNIX gave you no indication that the which allowed screen editing and that could stand improvement. previous character had been re-execution of previous com­ Much of what We implemented, or deleted when you pressed the mands. What we did was smaller asked to be allowed to implement, erase character. You had to accept in scope than what Bill later in­ is now a part of System V and 4.2 it on faith. This remains true on cluded in the C shell, but to Us it BSD; others of our innovations are many UNIX systems today. Most was unarguably better than what still missing from all versions of people on Cory Hall UNIX chang­ was then available. We ceased UNIX. Despite these accom­ ed their erase character to work on our projects only when it plishments, it seemed that backspace so that later characters became clear that Bill was whenever We asked The Powers would overwrite the erased ones, developing what would obviously That Be to install Our software but even that was not sufficient. become a new standard shell. Our and make it available to the rest of This was especially true when energies then were re-focused on the system's users, We were erasing a backslash, which persuading him to include our greeted with stony silence. counter-intuitively required two ideas. Some of our features Fred Brooks, in The Mythical erase characters. We wanted the ultimately were incorporated, Man-Month, describes the NIH system to show that the character some weren't. (Not Invented Here) Syndrome, Continued to Page 108 44 UNIX REVIEW JANUARY 1985 £•Jflfr l BERKELEY UNDERGROUND

When we finally checked our suspect me? Or was he just chat­ Continued from Page 44 handiwork the next day, we found ting? I could never tell whether was gone by blanking it out. We some serious flaws in the im­ Kurt was kidding; he had the most also wanted the line-erase plementation — not an uncom­ perfect poker face I had ever seen. character to display a blanked-out mon situation with work perform­ But he quickly made the question line. Some UNIX systems such as ed under extreme conditions. But academic, and proved again that 4.2 BSD and System V now sup­ the system was up and running, he was one of Them. port this, but it was not then and although the new features "I showed this to Bill, and he available anywhere under UNIX were flawed, they didn't seem to wanted to fix it", Kurt said. "Oh, Version 6. cause any problems, so we forgot really?" I stammered. "Sounds Bob and I had argued, about it for the time being. A week good to me," thinking that it was somewhat sleepily, for hours as to later, I was consulting in Cory — a real stroke of luck that Bill Joy the correct method of erasing we all offered free programming would be interested in the half- characters, and Bob had started help to other students in the time- completed project. If Bill finished putting our joint design into effect honored tradition of hackers it, then it would be in the system just as I collapsed on the floor for everywhere — when Kurt Schoens on legitimate grounds, and would "a short nap'*. I awoke around called me over to the other side of stay for good. dawn to find Bob asleep over the the room. Kurt paused for effect. "Yeah, terminal. When he woke up, he "Hey Doug," he said. "Look at he was all fired up about it, but I said he was pretty sure he'd finish­ this. It looks like someone tried to talked him out of it, and I just ed the job before falling asleep, but put character deletion into the ter­ deleted it from the system in­ neither of us had enough energy minal drivers, but only half finish­ stead." to check. It was time for food and ed." Oh, cruel fate! Kurt must 14 hours of sleep. My heart raced. Did he know that I was involved; he just UNIX* Marketing and Technical Professionals UNIFORUM - Dallas January 21-25 The Hottest UNIX* Opportunities Are With Gould's Firebreathing Team in Florida. Our family of 32-bit minicomputers blast the competition into oblivion. The Firebreather Team has the following opportunities: UNIX* Development Sales Support Marketing/Management • Drivers • Los Angeles • Product Market • Communications • Chicago • Market Development • Internals • Boston • Promotion • GOULD Electronics

Circle No. 63 on Inquiry Card

108 UNIX REVIEW JANUARY 1985 T\1 PC works

wanted to see me jump when he Dave couldn't help looking a little said "boo!" impressed; Kurt's "artificial in­ WORKS! Although I'm sure Kurt telligence" system was off to a thought the whole incident very great start. Kurt had talked to his Here at BASIS, we've funny, all I could think of was that budding mind for several minutes, checked FCworks carefully. yet another of my features had and Dave of course had grown And It does just what it gone down the drain. I discussed more and more impressed. Kurt this latest setback with others in and I faced the greatest challenge claims to do: it makes your the group, and we shared a sense of our lives in keeping a straight IBM PC™ (or compatible) of frustration. More than ever face during the demonstration, before, we were determined to get but we eventually made the a part of the UNIX™ our contributions accepted mistake of making the mind revolution. somehow. altogether TOO smart to be Kurt was both a graduate stu­ believable, in effect sending Dave dent and a system administrator, off to tackle more serious work. but I liked him all the same — There was one practical joke chiefly because of his practical that was notable for the length of jokes. We had recently cooperated time it was supported by the en­ in a spontaneous demonstration of tire group. The target was system Artificial Intelligence at the ex­ administrator Dave Mosher. Dave pense of an undergraduate named had been suspicious of bugs in our Dave who had joined Them as a system's homebrewed terminal system administrator. Dave had multiplexer for some time. Ross watched Kurt as he typed pwd to decided to persecute Dave by hav­ his shell prompt and received ing random characters appear on usr/kurt/mind as the response. his screen from time to time, His next command had been which of course convinced Dave mind -i -1 english . During all this that the terminal multiplexer did Now you and your PC can: time, Kurt was double-talking indeed have problems. To help about psychology and natural Ross with the prank, each of us • Access UNIX programs language processing and some sent Dave some garbage char­ • Transfer PC & UNIX files new approach to simulating the acters at random intervals human mind that he'd thought of. whenever any one of us was on the • Use the UNIX printer Dave looked dubious, but was will­ system. We had settled on the let­ ing to see how well Kurt's pro­ ter "Q" so that Dave would be sure • Back up PC files on UNIX gram worked. it was always the same bug show­ • Read, Create, & Send mail What Dave didn't realize was ing the same symptom. Since that Kurt had not been typing Dave had these problems no mat­ * Access public & private commands to the system at all; ter which terminal he was on, day networks although we were sitting not 10 or night, no matter who else was feet apart, Kurt and I had been logged onto the system, he was writing to each other and chatting positive there was a problem, and CALL or WRITE for for half an hour, and as a joke I he spent much time and effort try­ had been pretending I was Kurt's ing to get someone to fix it. More Information shell, sending him prompts and Unfortunately for Dave, he faking responses to commands. was the only one who ever saw fcfl.SJ.S. Dave had walked in at just the these symptoms, so everyone UNIX Software, Hardware. & Timesharing right time. So when Kurt typed thought he was a little paranoid. mind -i -1 english , I had natural­ We thought it was pretty funny at 1700 Shattuck #1 ly responded with: first, but after a few months of Berkeley, CA 94709 "Synthetic Cognition System, ver­ this, it seemed that Dave was real­ {415)841-1800 sion 17.8" ly getting rattled, so one day Ross PCworks i$ a trademark of Touchstone Software Interactive mode on, generated a capital "Q" as big as IBM PC is a trademark of IBM Corporation Language = english" the entire terminal screen and UNIX is a trademark of Bell Laboratories "Please enter desired conversational sent it to Dave's screen. This made topic: (default:philosophy)" it pretty obvious to poor Dave that Circle No. 64 on Inquiry Card

UNIX REVIEW JANUARY 1985 109 iftttei BERKELEY UNDERGROUND

someone, somehow, really had in the EECS department's PDP one system administrator would been persecuting him, and that he 11/70 in Cory Hall, since that was mail a new root password to all the wasn't paranoid after all. He had the original UNIX site and con­ other system administrators on an understandably low tolerance tinued to be the hotbed of UNIX the system, apparently not realiz­ for practical jokes after that. development, but We "collected" ing that we were monitoring their The numerous practical jokes all the other UNIX systems on mail for exactly this kind of securi­ we played were probably a reac­ campus, too. One peculiar aspect ty slip. Sadly, they soon guessed tion to the high level of stress we of the way the Underground had that this was not a good pro­ felt from our ongoing illicit opera­ to operate was that we rarely cedure, and we had to return to tions; it provided some moments knew the root password on functioning as "password-less of delightful release from what systems to which we had gained superusers", which at times could was, at times, a grim battle. There superuser access. This is because be a bit inconvenient. were many secret battles in the there were easier ways to get into, Late one night on Cory Hall war; if Our motto was "Features!", and stay into, a system than UNIX, as I was using my il­ Theirs was "Security for Securi­ guessing the root password. We legitimate superuser powers to ty's Sake" and the more the bet­ tampered, for instance, with the browse through protected but in­ ter. We were never sure how long su program so that it would make teresting portions of the system, I our victories would last; on the someone superuser when given happened to notice a suspicious- other hand. They were never sure our own secret password as well looking file called usr/adm/su. whether They had won. The war as when given the usual root This was suspicious because there lasted almost three years. password, which remained were almost never any new files in We were primarily interested unknown to us. In the early days, the administrative usr/adm direc­ tory. If I was suspicious when I saw the filename, I was half paralyzed when I saw it contain­ Great-looking TROFF output ed a full record of every command from low-cost laser printer! executed by anyone who had For several years, Textware has been licensing TPLUS software to process the worked as superuser since the output of TROFF and DITROFF for a wide variety of typesetters, laser printers, previous day, and I was in a full letter-quality and dot-matrix printers, and high-resolution displays. Our device- state of shock when I found, at the independent approach permits TPLUS to exploit the specific capabilities (and to end of the file, a record of all the work around the limitations) of all the imaging devices we support. commands that I'd executed dur­ Now we are excited to announce the availability of TPLUS for the Hewlett- Packard LaserJet printer. The low-cost LaserJet is a remarkable value on its ing my current surreptitious ses­ own—8 page per minute output speed, 300 dot per inch resolution, and sion, up to and including reading typesetter-quality fonts. TPLUS gives you access to all the damning file. 1 EQN examples | this and more from your own system. We support all the lim (tanx),in2x = 1 characters and accents needed by TROFF and EQN; in It took me perhaps 10 min­ 1 *-*»/2 1 addition, special characters (©; logos too) can be sup­ utes of panic-stricken worry before q+ff Sk k,k plied or generated to meet specific requirements. Our I realized that I could edit the sin(x) n< ' \ Jb>l precise handling of rules and boxes allows you to take record and delete all references to full advantage of TBL for forms, charts, etc. my illicit commands. I then im­ If you haven't been impressed with the output from low-cost laser printers in the past, take another look. While even LaserJet output is not in the same class as mediately logged out and warned the best of phototype, it is certainly well suited to documentation and a broad all other members of the group. range of other applications. When you do have a need for phototypeset images, Since nothing illicit ever appeared, TPLUS and the LaserJet will help save you time and money. Preview mode lets the system administrators were you proof all aspects of your documents conveniently, in-house, before sending lulled into a sense of false securi­ out for finished phototypesetting (from our UNKTEXT service). The HP LaserJet printer is not only inexpensive—it is an exceptional value! ty. Their strategy worked brilliant­ Want proof? This entire ad was set in position using TPLUS on the LaserJet! ly for us, allowing us to work in For further information, please write or call. peace for quite a while before the Also available for: next set of traps were laid. • AM 5810/5900 it 6400, APS 5 it fiS CG 8400 it 8600, Mergcnthaler 202 The next potential trap I • Xerox 2700, 8700 it 9700 TEXTWARE found was another new file in • BBN, Sun it 5620 CRT* INTERNATIONAL /usr/adm called password, that • Diablo, Qume it NEC LQPs PO Box 14 Harvard Square Telephone- • C Itoh it Epson dot-matrix kept track of all unsuccessful at­ Cambridge, MA 02238 (617) UNI-TEXT tempts to login as root or to su to root, and what password was used

HO UNIX REVIEW JANUARY 1985 Circle No. 65 on Inquiry Card Jf.fXtei BERKELEY UNDERGROUND

in the attempt. Since none of us took Us less than five minutes to always use that as a starting place. had known the root password for ascertain what the correct spelling We worked very hard to stay months and therefore weren't go­ was. one step ahead of Them, and we ing to become superuser by any­ One might think that, since spent most of our free time thing as obvious as logging in as we had several ways to become reading source code, in search of root, this wasn't particularly superuser anyway, it wouldn't either pure knowledge or another threatening to us, but it was very make any real difference whether weapon for the battle. At one time, interesting. The first few days that or not we knew the actual root I had modified every single utility we watched the file it showed password as well. The problem that ran as superuser with some attempts by legitimate system ad­ was that our methods worked on­ kind of hidden feature that could ministrators who had made ly so long as nothing drastically be triggered to give us superuser mistakes of various sorts. One of changed in the system; the usual powers. Chuck Haley once sent a Them once gave a password that way that They managed to win a letter to Jeff Schriebman com­ We discovered, through trial and battle was to backup the entire menting that he "had even found error, to be the root password on system from tape and recompile the card reader program" to show a different system. Several of all utilities. That sometimes set Us signs of tampering. I thought that Them gave passwords that seem­ back weeks, since it undid all of I had disguised it well, but it was ed to be the previous root our "backdoors" into superuser- extremely difficult to keep things password. Most of them were dom, forcing us to start from hidden from a group of system ad­ misspellings of the correct root ground zero on breaking into the ministrators who were not only password. Needless to say, this system again. But once we knew very intelligent, but also highly was a rather broad hint, and it the root password, we could knowledgeable about the inner UNIX™ VALIDATION

Validation Suite versions available

formerly • UNIX System III and System V(tmBeiiubs) THE • BSD 4.1 and BSD 4.2 SYSTEMS • /usr/group Standard GROUP • XENIX (tm Microsoft) PERENNIAL Validation Suites Include Source code for entire suite • Test driver control program Template for adding tests • System exercisor—stress tests Installation support service • Automatic and manual tests User documentation • Touch tests of all utilities Audit trail for all tests • Functional tests for all system calls

PERENNIAL SOFTWARE SERVICES GROUP 3130 De La Cruz Blvd. • Santa Clara, California 95054 • (408)727-2255 SOFTWARE DEVELOPMENT, TESTING, MAINTENANCE AND SUPPORT SERVICES

Circle No. 66 on Inquiry Card 112 UNIX REVIEW JANUARY 1985 UNDERGROUND

workings of UNIX. As an indica­ several months and through more changeover. Some were re- tion of the caliber of the people we than one system compilation implemented under Version 7 by were working against, I should before Dave Mosher noticed it (un­ those of the group who remained note that Chuck Haley is now a doubtedly with a sense of shock) at Berkeley, but by then many of researcher at Bell Labs; Bill Joy as he was patiently adding com­ us were leaving school, and the is VP of Engineering at Sun Micro­ ments to the previously un­ impetus behind our ideas left with systems; Kurt Schoens is a re­ documented kernel. us. searcher at IBM; Jeff Schriebman This sort of battling continued Ken Arnold is, perhaps, the is founder and President of for several years, and although most famous of our original group. UniSoft; and Bob Kridle, Vance They were suspicious of most of He stayed at Berkeley longer than Vaughn, and Ed Gould are Us at one time or another, none of any of the rest of us, and became founders of Mt. Xinu. Us was ever caught red-handed. It well known for such contributions This was an unusual situa­ undoubtedly helped that we never as Termlib, curses,fortune, Mille tion; system administrators are performed any malicious acts. We Bourne, and of course his co- not usually this talented. Other­ perhaps flouted authority, but we authorship of Rogue. But some­ wise, they'd be doing software always enhanced the system's how it seemed a Pyrrhic victory development rather than admin­ features. We never interfered with even for Ken; much of his best istration. But at the time, there the system's normal operation, work in the early years never saw was no one else capable of doing nor damaged any user's files. We the light of day. UNIX system administration. learned that absolute power need We could not help but feel As a result, we had to move not corrupt absolutely; instead it that we had passed through a sort quickly, quietly, and cleverly to taught us restraint. of Dark Age for UNIX develop­ stay ahead, and planting devious This is probably why we were ment, and even with the devices in the midst of standard eventually accepted as members Renaissance in full bloom, We software was our primary tech­ of the system staff, even though ponder what might have been, nique. Normally trusted programs by then several of Us had confess­ and bewail the features that UNIX which have been corrupted in this ed to our nefarious deeds. Once we will now never have. way are called "Trojan Horses", were given license to modify and Doug Merritt became one of the after the legend of the Greeks who improve UNIX, we lost all motiva­ earliest UNIX users outside Bell were taken in by a bit of misplaced tion to crack system security. We Laboratories while attending UC trust. One of our favorite tricks for didn't know it at the time, but this Berkeley in 1976. He helped to debug hiding our tracks when we modi­ has long been known to be one of termcap and contributed to the fied standard utilities was the the most effective ways of dealing development of vi and curses. Mr. diddlei program, which allowed with security problems: hire the Merritt now works as a consultant us to reset the last change time on offenders, so that there is no more in the San Francisco Bay Area. a modified file so that it appeared Us versus Them, but simply Us. Bob Toxen is a member of the to have been unchanged since the It worked well in our case; technical staff at Silicon Graphics, previous year. Bob modified the under the auspices of the System Inc., who has gained a reputation as setuid system call in the UNIX Development and Research a leading expert on uucp com­ kernel so that, under certain cir­ Group, created by the ever- munications, file system repair and cumstances, it would give the pro­ industrious Dave Mosher, we went UNIX utilities. He has also done gram that used it root privileges. happily to work on UNIX develop­ ports of System III and System V The "certain circumstances'* con­ ment. The development of UNIX to systems based on the Zilog 8000 sisted simply of leaving a capital at Berkeley, always fast-paced, and Motorola 68010 chips. "S" (for Superuser) in one of the exploded once everyone — in­ Best known as the author of machine's registers. Bob was bold cluding undergraduates — were curses and co-author of Rogue, Ken enough to leave this little feature participating. Arnold was also President of the in the system's source code. We The only fly in the ointment Berkeley Computer Club and the usually put our Trojan Horses in was the introduction a short while Computer Science Undergraduates the system executables only — to later of UNIX Version 7. While it Association during his years at UC decrease the chance of it being was a vast improvement in many Berkeley. He currently works as a noticed. But Bob took the chance ways over the Version 6 that we programmer in the Computer so that the feature would persist had been working with, most of Graphics Lab at UC San Francisco even if the system were recompil­ the enhancements we had and serves as a member of the UNIX ed. Sure enough, it lasted for developed were lost in the REVIEW Software Review Board. • 114 UNIX REVIEW JANUARY 1985 $ HE GENESIS STORY

;vc , An unofficial, ^irreverent, incomplete account of how * the UNIX operating system developed --•..--•••-.-

*' -^ -. . by August Mohr ^;^r

*afeg|

ls^J:>^i?>^j"wif!^^5^,5VT^ ^

-•3SSJ%KS(*i»

^•i^^^^^^^^r-' ^Aftr>^V; Jiiii^nnjM^Ji^B.^jiLjj^.mji 3 rs. *-a*-:* «>v«v%4. ...

i^&:-"Z<

|&^^ ^yL **1gg^^

his is. so to speak, a history of how UNIX cvolv- the idea that Bell Laboratories and .Western Electric- edasapro^uei: notthe*vofficiar history of who had engaged in a masterful piece of long-term • was responsible, for WTbar features, and what strategy by first releasing UNIX to universities and year which tnjle^tpnes were crossed, but the then later releasing it to the commercial world..How- -mpblihcarV^^°iy.°',h0^* decisions were made and clever. I thought, to get universities involved in the whatmotivated tffe people involved. Most of the development of the system and gel a whole erop of . readers^of'this magazine are familiar with the system UNIX experts trained in the process. But after talk­ itself, so I don't want to go into great detail about how ing to some of the people involved. I have reversed the system goVtobewhat j.t is internally-Jutt rather my opinion. Bell. it seems, was dragged kicking and how it"g6rto be at all. ;? "\ :>\ screamihi! into providing UNIX to the world...... -p^p^ years ago. I u^s one of niany espousing

Illustration bv firth' Dun GENESIS STORY

IN THE BEGINNING, THERE Ritchie's Murray Hill office — was cation), had his first exposure to WAS MULTICS looking for a word processing UNIX at about this time. As he Multics, in some ways UNIX's system at about this time. A remembers it, "I came across Ken predecessor, was a huge project, a paper-tape system for an old Thompson by accident and wanted combined effort of three of the Teletype machine was under to do some software development largest computing centers of the consideration. on a PDP-11. He said he had a day. All three principles — Bell, With a bit of salesmanship, machine I could use but that it GE, and MIT, had already done Thompson and Ritchie got the didn't run on a DEC operating operating systems, even time- legal team interested in a UNIX- system. I had lots of exposure to shared systems, before, and supported system. It proved to be timesharing systems and it was following Fred Brooks' Second pretty obvious that this was System Syndrome, it was felt superior in many ways. Multics was going to solve all the "It was uphill in the begin­ problems of its predecessors. ning. The fact that it was Ultimately, the project proved to The combination of homegrown in the Labs didn't cut be too late and Bell dropped out, anything with the people that I leaving Ken Thompson, Dennis program was working with at that time. Ritchie, and Rudd Canaday with­ They went and visited Ken and out a timeshared system to play development and Dennis in their sixth-floor attic at with. Murray Hill and all they saw was So they set about building an word processing hardware laying all over the floor operating system of their own. and a bunch of guys in T-shirts So until this point, time­ was to have and sneakers. It wasn't the sort of sharing systems had only been place that would warm the heart developed for big machines serendipitous of a manager who had been costing hundreds of thousands brought up in a traditional data of dollars. It was unlikely that effects. processing environment. It really Computer Research was going to was the *two good guys in a get another investment of that sort garage* kind of syndrome except out of Bell so soon after Multics. that they happened to be in an In any event, Thompson, Ritchie, attic. and Canaday weren't particularly a momentous deal. The UNIX proj­ "That first time I met Ken interested in working on another ect got the machine it needed and Thompson, he had a small book­ large scale project, so they set the legal department got the word shelf, maybe 2 1/2 feet long, above their sights on obtaining a processing system it wanted — an old Teletype Model 37 terminal. minicomputer on which to build a along with some rather impressive On the shelf were hardware timesharing system they could local support. manuals from Digital Equipment use as a program development en­ A number of important bar­ for the PDP-11. occupying maybe vironment. They wanted the kind riers thus were crashed. First, the five inches, and the rest of the of flexibility and power they had business side of Bell Labs, general­ shelf was nothing but chess worked toward in Multics, without ly held at arm's length by the peo­ books. I think Thompson will tell incurring the expensive rings of ple in Research, got a healthy dose you himself that he developed protection and interlocks. of Research assistance. More im­ UNIX as a good place to develop The project was short on portantly for UNIX users, Thomp­ chess programs. It turns out that computing power, though. With son and Ritchie got their first live it's everybody else's idea of where only a PDP-7 to work on, the customer. UNIX word processing and how to develop programs researchers yearned for another suddenly had to be usable by too." machine — preferably a VAX secretarial staff as well as by pro­ Haight has had many years of 11/20. To get the necessary funds gramming staff. The changes experience in the Bell Labs for a new machine, though, it was made to accomplish this transi­ environment. Before moving to clear they would first need to find tion were to have serendipitous his current post, he served as an application. effects as UNIX evolved. part of the UNIX Support group Fortunately, the Bell Labs' Richard Haight, now AT&T and the PWB (Programmer's Legal Department — which was Bell Labs Supervisor of Video Workbench) development group. close at hand to Thompson and Systems Software (a UNIX appli­ He believes that the environment 20 UNIX REVIEW JANUARY 1985 GENESIS STORY

of the Labs was a major factor in this purpose for the system as a and applications software back­ the creation of UNIX. whole. grounds who were considering "There's a small fraction of Doug Mcllroy is usually credi­ writing their own- operating people who just go crazy over ted with the idea of adding pipes, system — their first — when computers," he explained. We although the idea may have been Berkley Tague, now head of Bell hire a bunch of them and they around since Multics. Haight Labs' Computing Technology remain workaholics for several believes there may have been yet Department, suggested they use years after coming in. Eventually another reason for implementing UNIX to get started. they get a house and a mortgage them. The original UNIX had a file t4I observed that people were and they get married and have size limitation of 64K and, accord­ starting to put these minis out in kids and settle down to be normal ing to Haight. "one of the people the operating company, and saw human beings, but it's wonderful there in research was constantly that it was an area of both oppor­ to hire these people because you blowing that size restriction tunity and potential problems," get two or three people's work out in an intermediate pass of the Tague remembered. "I found that of them for several years. Those Assembler." Between Mcllroy's some of the people in development are the kind of people that give lobbying for the idea and this had never built an operating you things like UNIX." other problem with file size, system for any computer before; Thompson and Ritchie were final­ many of them had very little soft­ AN INFLUENTIAL FEATURE: ly convinced to implement pipes. ware background. They were PIPES coming out of hardware develop­ Dick Haight is one of the MOVING INTO THE COMPANY ment and telephone technology people who became hooked on Independent of what was hap­ backgrounds, and yet were starting UNIX at the very beginning, and pening in the research area, Bell to build their own operating even yet is an outspoken pro­ was starting to perceive the need systems. Having been through ponent of its value and power — that phase of the business myself, although he will tell you he is still it seemed silly to go through it waiting for something better to another hundred times, so I come along. "There's a lot of com­ started pushing the UNIX plaint about UNIX being terse and "Naturally I knew operating system into these pro­ for experts only," he said. "I've jects.'* seen that blamed on the pipe that once they got Tague's backing of UNIX, as mechanism. If terseness is the a development system for opera­ price for having pipes, I'll take it on UNIX they tions, was not just a personal any day." preference. "I had every con­ As one of the new system's wouldn't be able to fidence in the people who built it first users, Haight got to follow because I'd worked with them on many of the changes. "I happened get off. It's just like Multics," he explained. "With to have been visiting the research their experience and training, I crew the day they implemented drugs." figured they could build a much pipes," he recalled. It was clear to better operating system than everyone, practically minutes somebody who's building one for after the system came up with the first time, no matter how pipes working, that it was a for minicomputer support for its smart that person is." wonderful thing. Nobody would telephone operations. It needed ever go back and give that up if Operations Systems, not Oper­ SUPPORT? they could help it." ating Systems. With the numbers UNIX had been running long Haight believes that the pipe of systems under consideration, enough in research by that time facility has had a major effect on the possibility of being tied to a that Tague knew that the system the evolution of UNIX. Because of single vendor, or having each site the operations group would get the shell and the piping facility, tied to a different vendor, induced would serve as a very good start­ programs running under UNIX a kind of paranoia. There just had ing point. Unfortunately, there can be conversational without be­ to be another way. was no vendor support for it. ing explicitly programmed that The groups responsible for The argument Tague made way. This is because piping adds developing operations systems for UNIX was: if the operations to the shell's ability to accomplish had many people from hardware people were going to build their 22 UNIX REVIEW JANUARY 1985 KEY EVENTS IN UNIX SYSTEM EVOLUTION

- EDUCATIONAL CACIMf I COMMERCIA1 L —z z 1 ? A 4 A TMOTOROL A MARKETING LICENSES ARTICLCACM[COMME I LICENSEI S INTERACTIVE C BOOK MICROSOFT I ATAT I IBM PC/IX NSYSTE M V/88 MARKETS I SUPPORT! DEC ULTR1X S TICLE I LICEN! SYSTEMS UNIX BSTS XENIX XENIX " • MARKETS UNISOFT MICRO DISTRIBUTION IS/1 MARKETS CONTRACTS TOBOC UN1PLUS BELL LABS "1 X 1 TT RESEARCH x DATAKIT PDP-11 V8-UNIX MERT/UNIX V-7 PORTABLE BLIT WRITTEN IN C UNIX ON 32-V-UNIX V8-DIST OS DEVELOPMENT INTERDATA 8/32 ON VAX NETWORK ARCH —x— ANNUAL PWB RELEASES PWB MERT CB-UNIX UNIX USE BY TO SUPPORT 10 UNIX (MERT 370.1100 BELL LABS DEV DEPTS DEV DEPTS DROPPED) t? SUPPORT T X "X SUPPORT BEGINS T X RELEASE SYSTEM RELEASE SYSTEM SYSTEM V 3 III 4 V RELEASE 2 RELEASE I

—x x x x—x—i" THOMPSON VERSION 2.0 VERSIOON 4.0 I 4 1C 4.2 BERKELEY SABBATICAL (11/70) (VAX) RESEARCH VERSION 4.1 J_J I I I I I L J I I I L '70 *72 '74 '76 '78 '80 '82 *84 Courtesy of Larry Crume, AT AT UNIX Pacific. own system, they were going to network of minicomputers would worked closely with Bell Labs have to maintain it themselves. be a significantly different prob­ Research, its concerns sometimes Surely, UNIX could be no worse. lem, though. Maintaining an diverged. They could use it to get started operating system is not at all like One area the two groups could and do the development. If a more maintaining an electronic switch­ agree on, though, was portability. efficient or better operating ing system. The minicomputers By 1973, it was already on the system was needed for a target had different reliability demands, horizon. Tague foresaw the machine when they got into the requiring a different support possibility of UNIX becoming an field, they could always build it, structure in the organization — interface between hardware and but UNIX would at least get them one that did not yet exist in any software that would allow applica­ off the ground. "Naturally I knew form. In many ways, the opera­ tions to keep running while the that once they got on this thing tions group was breaking new hardware underneath was they wouldn't be able to get off. ground. changing. It's just like drugs," Tague Up to this point, Tague had From the support point of explained. served as head of the Computer view, such a capability would Tague also knew it was impor­ Planning department responsible solve a very important problem. tant to get some field support. for systems engineering. After Without UNIX and its potential "We were starting to put these gaining support for UNIX in the portability, the people building things in the operating companies operations group over the course the operations support systems all around the countryside and the of 1971 and 1972, he made a push were faced with selecting an out­ prospects were that there were for two significant changes. The side vendor that could supply the going to be several hundred minis first was to make UNIX an inter­ hardware on .which to get their over the next few years that were nal standard and the second was development done. Once that was going to have to be maintained to offer central support through complete, they would be locked with all their software and hard­ his organization. In September, into that vendor. Portability ob­ ware," he said. 1973, he was permitted to form a viated this limitation and offered Bell had already gained some group called UNIX Development a number of other advantages. field support experience maintain­ Support, the first development When making a hardware upgrade, ing electronic switching machines organization supporting a "Stand­ even to equipment from the same and their software. Supporting a ard UNIX". While this group vendor, there are variations from 24 UNIX REVIEW JANUARY 1985 GENESIS STORY

version to version. That could cost and a suitable compromise was Bell Laboratories because people a lot of money in software revi­ made. "As one of the local gurus loved to use it," he said. sions unless there were some level put it to me one time," Tague said, Canaday first experienced of portability already written into "one of the problems in UNIX is UNIX, although it hadn't yet been the scenario. Fortunately, the in­ that everybody wants to carve his named, while working with tegral portability of the system initials in the tree." When the Thompson and Ritchie. He left developed by Research proved choice came down essentially to that group and later became head adequate to make UNIX portable of a group building large, over a wide range of hardware. mainframe-oriented systems. The first UNIX applications Because of his previous exposure were installed in 1973 on a system to UNIX, he wanted to bring it to involved in updating directory in­ From the very his new group. formation and intercepting calls to Canaday found lots of support numbers that had been changed. beginning within among the programmers who had The automatic intercept system already tasted UNIX. One of those, was delivered for use on early Bell, UNIX followed Richard Haight, recalled, "I was PDP-1 Is. This was essentially the trying to get my management to first time UNIX was used to sup­ what has become a get UNIX and we dreamed up the port an actual, ongoing operating idea of using it as a common business. familiar pattern of timesharing interface to different To Tague, at this time, "our kinds of host computers." real problem was pruning the users leading their Initially, the project involved tree". There were so many dif­ interfacing with big IBM and ferent sites using UNIX that each management. Univac machines, and later ex­ would come up with different panded to interfacing with RCA, answers to the same problems of Xerox, and others. The basic idea printer spooling, mail, help, and so was to edit programs and work on. 'The customers would invent with files under UNIX, but instead this stuff and make it work and tossing a coin, Tague, as arbirator, of compiling and executing under our problem was to get the slight­ tried to make sure that each group UNIX, you could send the remote ly different variations together, get got at least one pet contribution job off to a big machine. This way, the best of all of those worlds, put into the system. the programmer didn't have to it in the standard, and get it out Fortunately, UNIX is flexible deal with complicated IBM JCL se­ again." This was, in many ways, enough that even the particularly quences since he could just give a political process. Tague credits traumatic decisions, such as the the UNIX utility the parameters it the "technical underpinnings" for ones concerning standard shell needed to know. The masses of making the process easier than ex­ versions, could be patched in printout could then come back as pected. "That made it easy to get slowly — at the user's discretion. a file under UNIX and, as Dick the right stuff in without upsetting Haight put it, "save cutting down the whole world. I didn't have to A FAMILIAR PATTERN a tree." It also saved having to go to all of my customers and tell From the very beginning retrain programmers for a variety them that this was now my new within Bell, UNIX followed what of host systems. version and that nothing they had has become a familiar pattern of This original Programmer's out in the field would run again," users leading their management. Workbench system was built on a he said. While this is riot the most com­ PDP 11/45. The system eventually To provide a standard UNIX mon marketing strategy in the offered lots of utilities, including system, the support group had to commercial world, it is typical of ones for analyzing host machine establish what version it would Bell Labs' "bottom-up" organiza­ dumps on the UNIX system. back. This was a process of tion. According to Rudd Canaday, While work proceeded on the negotiation and compromise with now head of Bell Labs' Artificial PWB system, an interesting dis­ the UNIX-using community — not Intelligence and Computing En­ covery was made. The designers a unilateral decision. The support vironments Research Department, had assumed that the majority of team and customers often ended change within the Labs often the work cycle would involve the up arguing things out until comes from the people doing the host computer. Users were thought everybody understood the issues work. "UNIX spread throughout of as editing a file, sending it to a 26 UNIX REVIEW JANUARY 1985 JtJflfc l RESEARCH ENVIRONMENT

host computer, getting the print them. This happened, according convinced. Canaday said to file back, looking at it. and doing to Haight. because programmers management, "Well, if you want that over and over again. As it have to be able to document pro­ to keep on using this, you're going turned out. samples taken from grams on the same machine used to have to start buying machines different kinds of work groups on for development. "Things like to do it." He knew that "once you different systems showed people pipes and the power of the shell let people get their hands on tended to use the text formatter. are not to be slighted, but what's UNIX, they just won't let go." nroff. five times as often as they really important is the fact that A key piece in the rapid submitted Remote Job Entry you can do your documentation spread of UNIX within Bell Labs programs. and your programming on the was the low price of mini­ This unexpected result might same machine." he said. You can computers relative to mainframes. not have happened had UNIX not be editing your documentation A department head's urging was had fairly sophisticated word pro­ and break away from that to edit generally sufficient for purchase of cessing facilities available to the source. When you're finished a VAX. Mainframe purchases were programmers. The original devel­ with that, you can submit a com­ considerably more sticky. A VAX opment for Bell's legal department pile in the background and go had sufficient power to reasonably could hardly be called "incredible back to editing your documenta­ serve the needs of a department, foresight", but happily for UNIX, tion while the compile happens." so VAXen became increasingly word processing was to become Flushed with the success of commonplace. the single most commonly used the PWB and the Remote Job En­ More and more departments computer application. Once the try facility, Canaday and his group were becoming convinced that facilities were there, programmers set about showing people what UNIX was part of the path toward made massive, unexpected use of was possible. Once the users were Continued to Page 117 REFLECTIONS ON SOFTWARE RESEARCH

Can the circumstances that nurtured the UNIX projectJ be produced again? ^

by Dennis M. Ritchie

The UNIX operating system system; Joe Ossanna, Doug the ACM the next year has suddenly become news, but it Mcllroy, and Bob Morris were es­ * Thereafter, Its ufce grew is not new. It began in 1969 when pecially enthusiastic critics and steadily, both inside and outside of Ken Thompson discovered a little- contributors. In 1971, we acquired Bell Laboratories. A development used PDP-7 computer and set a PDP-1 1, and by the end of that group was established to support out to fashion a computing en­ year we were supporting our first projects inside the company, and vironment that he liked. His work real users: three typists entering several research versions were soon attracted me; I joined in patent applications. In 1973, the licensed for outoMe'.twc:'^^/""./ the enterprise, though most of the system was rewritten in the C The last research distribution ideas, and most of the work for language, and in that year, too, ft wad the seventh edition system, that matter, were his. Before long, was first described publicly at the which appeared tit 1979; more others from our group in the Operating Systems Principles con­ recently, AT&T began to market research area of AT&T Bell ference; the resulting paper J8J System HI, and now offera System Laboratories were using the appeared in Communications of *•-?'' 28 UNIX REVIEW JANUARY 1985 Jf ftk\ RESEARCH ENVIRONMENT

Continued from Page 28 that UNIX was created, we were UNIX* and C V, both products of the develop­ pushing hard for a machine, either TRAINING ment group. All research versions a DEC PDP-10 or SDS (later Xerox) were "as is," unsupported soft­ Sigma 7. It is certain, in retro­ UNIX Overview (1) $145 spect, that if we had succeeded in UNIX Fundamentals (3 days) $595 ware; System V is a supported C Fundamentals (3 days) $595 product on several different hard­ acquiring such a machine, UNIX Austin Schedule ware lines, most recently in­ might have been written but Feb. 6-8 UNIX cluding the 3B systems designed would have withered away. Feb. 13-15 C Similarly, UNIX owes much to Mar. 9-11 UNIX and built by AT&T. Mar. 16-18 c UNIX is in wide use, and is Multics [5], I have described On-site training available now even spoken of as a possible [6, 7], it eclipsed its parent as much because it does not demand Call or write to register or to get on our mailing industry standard. How did it list. come to succeed? unusual hardware support as TELOS CONSULTING SERVICES There are, of course, its because of any other qualities. 815 Brazos, Suite 504 technical merits. Because the Finally, UNIX enjoyed an Austin, TX 78701 system and its history have been unusually long gestation period. 512/478-2072 discussed at some length in the During much of this time (say •UNIX is a trademark of Bell Laboratories. literature [6, 7, 11], I will not talk 1969-1979), the system was effec­ about these qualities except for tively under the control of its one. Despite its frequent surface designers and being used by inconsistency, so colorfully an­ them. It took time to develop notated by Don Norman in his all of the ideas and software, but Datamation article [4] and despite even though the system was still its richness, UNIX is a simple, being developed, people were coherent system that pushes a few using it, both inside Bell Labs, good ideas and models to the limit. and outside under license. Thus, It is this aspect of the system, we managed to keep the central above all, that endears it to its ideas in hand, while accumulating adherents. a base of enthusiastic, technically Z Systems is now offering its port Beyond technical considera­ competent users who contributed of the Berkeley cshell for your ideas and programs in a calm, AT&T 3B2. Through enrollment tions, there were sociological in the 3B Subscription Service, forces that contributed to its suc­ communicative, and noncompeti­ you will get the first software cess. First, it appeared at a time tive environment. Some outside release which includes: when alternatives to large, cen­ contributions were substantial, • Berkeley 4.2 compatible such as those from the University • csh (less jobs) trally administered computation • Is (links to lf.lr, . . .) centers were becoming possible; of California at Berkeley. Our • more the 1970s were the decade of the users were widely, though thinly, • strings minicomputer. Small groups distributed within the company, • and others could set up their own computa­ at universities, and at some Your annual subscription to commercial and government 3BSS also includes newsletters tional facilities. Because they were and future software releases for starting afresh, and because organizations. The system be­ a nominal media and handling manufacturers' software was, at came important in the intellec­ charge. best, unimaginative and often tual, if not yet commercial, The cost is just $200/yr. horrible, some adventuresome marketplace because of this net­ work of early users. To order call 303-526-1633 people were willing to take a major credit cards accepted chance on a new and intriguing, What does industrial compu­ even though unsupported, ter science research consist of? operating system. Z SYSTEMS, INC Some people have the impression Second, UNIX was first 602 Park Point Drive that the original UNIX work was Genesee Business Center available on the PDP-11, one of a bootleg project, a "skunk Golden, CO 80401 the most successful of the new works". This is not so. Research minicomputers that appeared in workers aire supposed to discover Call |or information on gtj^r ;•• soft ware and hardware^rQdvicts? the 1970s, and soon its portability or invent new things, and brought it to many new machines although in the early days we Circle No. 70 on Inquiry Card as they appeared. At the time subsisted on meager hardware, 118 UNIX REVIEW JANUARY 1985 we always had management en­ port in order for the company to nounce the name). In spite of the couragement. At the same time, make best use of them. In the hashing through 12 buttons it was certainly nothing like a past, this use would have been (which, for example, squashed development project. Our intent exclusively inside the Bell System; "A", **B" and "C" together), it was was to create a pleasant com­ more recently, there is the possi­ acceptably accurate: it had to give puting environment for ourselves, bility of developing a product for up on around 5 percent of the and our hope was that others direct sale. tries. The program was a local hit liked it. For example, some years ago and well-used. Unfortunately, we The Computing Science Mike Lesk developed an auto­ couldn't find anyone to take it Research Center at Bell Lab­ mated directory-assistance over, even as a supported service oratories to which Thompson and system [3]. The program had an within the company, let alone a I belong studies three broad areas: online Bell Labs phone book, and public offering, and it was an ex­ theory; numerical analysis; and was connected to a voice syn­ cessive drain on our resources, so system, languages, and software. thesizer on a telephone line with it was finally scrapped. (I chose Although work for its own sake a tone decoder. One dialed the this example not only because it resulting, for example, in a paper system, and keyed in a name and is old enough not to exacerbate in a learned journal, is not only location code on the telephone's any current squabbles, but also tolerated but welcomed, there is key pad; it spoke back the per­ because it is timely: the organiza­ strong though wonderfully subtle son's telephone number and office tion that publishes the company pressure to think about problems address (it didn't attempt to pro­ telephone directory recently asked somehow relevant to our corpora­ tion. This has been so since I join­ ed Bell Labs around 15 years ago, and it should not be surprising, the old Bell System may have seemed a sheltered monopoly, but research has always had to pay its way. Indeed, researchers love to find problems to work on; one of the advantages of doing research in a large company is the enor­ mous range of the puzzles that ...enough said. turn up. For example, theorists may contribute to compiler design, or to LSI algorithms; numerical analysts study charge and current distribution in semiconductors; and, of course, software types like to design systems and write programs that people use. Thus, computer research at Bell Labs has always had considerable commitment to the world, and does not fear edicts commanding us to be practical. Saa us at: Fulcrum Technologies Inc. UniForum 331 Cooper Street, For some of us, in fact, a prin­ Dallas Ottawa, Canada cipal frustration has been the Jan. 21 to 25 K2PQG5 inability to convince others Booth 2959 (6131238-1761 that our research products can indeed be useful. Someone may invent a new application, write an •# m w mj mu wt illustrative program, and put it to wi m mm i w mii rj use in our own lab. Many such demonstrations require further development and continuing sup­ Circle No. 71 on Inquiry Card UNIX REVIEW JANUARY 1985 119 * itki RESEARCH ENVIRONMENT

us whether the system could be Laboratories; one report in to argue, is that of irrelevance to revived.) Science [2] is typical. One fear the goals of the company. Com­ Of course not every idea is sometimes expressed is that basic puter science research is different worth developing or supporting. research, in general, may languish from these more traditional In any event, the world is because it yields insufficient short- disciplines. Philosophically it changing: our ideas and advice are term gains to the new, smaller differs from the physical sciences being sought much more avidly AT&T. The public position of the because it seeks not to discover, than before. This increase in company is reassuring: moreover, explain, or exploit the natural influence has been going on for research management at Bell world, but instead to study the several years, partly because of Labs seems to believe deeply, properties of machines of human the success of UNIX, but more and argues persuasively, that the creation. In this it is analogous to recently, because of the dramatic commitment of support to basic mathematics, and indeed the alteration of the structure of our research is deep and will continue "science" part of computer company. [1]. science is, for the most part, AT&T divested its telephone Fundamental research at Bell mathematical in spirit. But an operating companies at the be­ Labs in physics, chemistry, and inevitable aspect of computer ginning of 1984. There has been mathematics may, indeed, not be science is the creation of computer considerable public speculation threatened; nevertheless, the programs: objects that, though about what this will mean for danger it might face, and the case intangible, are subject to commer­ fundamental research at Bell against which it must be prepared cial exchange. More than anything else, the greatest danger to good computer science research today may be excessive relevance. Evidence for UNICOMP the worldwide fascination with computers is everywhere, from Technical Type UNIX*/ C the articles on the financial, and even the front pages of the news­ Typesetting Service $25,000 papers, to the difficulties that even to the most prestigious universities experience in finding and keeping • UNIX*, Troff, Wizard $50,000 faculty in computer science. The • Compugraphic 8400 Typesetter best professors, instead of APPLICATIONS teaching bright students, join • Quality appearance of books, start-up companies, and often proceedings, newsletters. and SYSTEMS discover that their brightest students have preceded them. • Specializing in National Placement Leader Computer science is in the technical documents. for UNIX/C PROF'Ls limelight, especially those aspects, • Documents accepted via such as systems, languages, and machine architecture, that may magnetic tape, 500 OPENINGS NATIONWIDE have immediate commercial appli­ phone lines, or paper. Call or send resume cations. The attention is flattering, • For information, samples, but it can work to the detriment of or estimates, call J. ROGERS ASSOCIATES good research. Dept. CR As the intensity of research in 505/662-EDIT (3348) 123 Franklin Corner Rd. a particular area increases, so does Lawrenceville, N.J. 08648 the impulse to keep its results secret. This is true even in the 1580 Camino Redondo (609) 896-8484 university (Watson's account [12] Los Alamos, NM 87544 (800) 2220596 Representing of the discovery of the structure EOE Companies of DNA provides a well-known ex­ •Unix is a trademark of Bell Laboratories 'UNIX is a trademark of Bell Labs ample), although in academia there is a strong counterpressure: Circle No. 73 on Inquiry Card Circle No. 74 on Inquiry Card unless one publishes, one never 120 UNIX REVIEW JANUARY 1985 tfttei ENVIRONMENT

becomes known at all. In industry, On the other hand, in smaller and workers will do well to a natural impulse of the establish­ companies, even the most keep in mind how, and under ment is to guard proprietary infor­ vigorous research support is what conditions, the UNIX system mation. Researchers understand highly dependent on market con­ succeeded. If we can keep alive reasonable restrictions on what ditions. The New York Times, in enough openness to new ideas, and when they publish, but many an article describing Alan Kay's enough freedom of communica­ will become irritated and flee passage from Atari to Apple, notes tion, enough patience to cdlow elsewhere, or start working in less the problem: "Mr. Kay...said that the novel to prosper, it will remain delicate areas, if prevented from Atari's laboratories had lost some possible for a future Ken Thomp­ communicating their discoveries of the atmosphere of innovation son to find a little-used CRAY/I and inventions in suitable fashion. that once attracted some of the computer and fashion a system as Research management at Bell Labs finest talent in the industry. creative, and as influential, as has traditionally been sensitive to "When I left last month it was UNIX. maintaining a careful balance be­ clear that they would be putting tween company interest and the their efforts in the short term," REFERENCES industrial equivalent of academic he said..."I guess the tree of 1. Bell Labs: New order augurs well. freedom. The entrance of AT&T research must from time to time Nature 305, 5933 (Sept. 29, 1983). into the computer industry will be refreshed with the blood of 2. Bell Labs on the brink. Science 221 test, and perhaps strain, this bean counters."[9] (Sept. 23, 1983). balance. Partly because they are new 3. Lesk, M.E. User-activated BTL direc­ Another danger is that com­ and still immature, and partly tory assistance. Bell Laboratories inter­ mercial pressure of one sort or because they are a creation of the nal memorandum (1972). another will divert the attention of intellect, the arts and sciences 4. Norman, D.A. The truth about UNIX. the best thinkers from real innova­ of software abridge the chain, Datamation 27(1981). tion to exploitation of the current usually in physics and engineer­ 5. Organick, E.I. The Multics System MIT fad, from prospecting to mining a ing, between fundamental dis­ Press Cambridge. M.A. (1972) known lode. These pressures mani­ coveries, advanced development, 6. Ritchie, D. M. UNIX time-sharing fest themselves not only in and application. The inventors of system: A Restrospective. Bell Syst. the disappearance of faculty into ideas about how software should Tech. J. 57.6 (1978) (1947-1969. industry, but also in the con­ work usually find it necessary to 7. Ritchie, D. M. The Evolution of the servatism that overtakes those build demonstration systems. For UNIX time-sharing system. In Language with well-paying investments — large systems, and for revolu­ Design and Programming Methodology. intellectual or financial — in a tionary ideas, much time is Jeffrey M. Tobias, Ed. Springer-Verlag, given idea. Perhaps this effect required. It can be said that UNIX New York. (1980). explains why so few interesting was written in the '70s to distill 8. Ritchie, D. M. and Thompson, K. The software systems have come from the best systems ideas of the '60s, UNIX time-sharing system. Commun. the large computer companies: and became the commonplace of ACM 17.7 (July 1974) 365-375. they are locked into the existing the '80s. The work at Xerox PARC 9. Sanger, D. E. Key Atari scientist swit­ world. Even IBM, which supports on personal computers, bitmap ches to Apple. The New York Times a well-regarded and productive graphics, and programming en­ 133,46,033 (May 3, 1984). research establishment, has in vironments [10] shows a similar 10. Thacker, C. P. et al. Alto, a personal recent years produced little to progression, starting and coming computer. Xerox PARC Technical Report cause even a minor revolution in to fruition a few years later. CSL-79-11. the way people think about Time, and a commitment to the 11. Thompson, K. UNIX time-sharing computers. The working ex­ long-term value of the research, system: UNIX implementation. Bell Syst. amples of important new systems are needed on the part of both Tech J. 57,6 (1978). 1931-1946. seem to have come either from the researchers and their 12. Watson. J. D. The Double Helix: A entrepreneurial efforts (VisiCalc is management. Personal Account of the Discovery of the a good example) or from large Bell Labs has provided this Structure of DNA. Atheneum Publishers. companies, like Bell Labs and commitment and more: a rare New York (1968). most especially Xerox, that were and uniquely stimulating research Copyright 1984 by Association much involved with computers environment for my colleagues for Computing Machinery, reprinted and could afford research into and me. As it enters what com­ with permission of Communications them, but did not regard them as pany publications call "the new of the ACM, August 1984, Volume their primary business. competitive era", its managers 27, Number 8M 122 UNIX REVIEW JANUARY 1985 UNIX UNLEASHED

The university role of research in maintaining system vitality

by Marshall Kirk McKusick

SF ince the AT&T divestiture, orderly system evolution. Be that UNIX has become the focus of a as it may, AT&T's major UNIX massive marketing effort. To suc­ commercialization effort has ceed, this effort must convince succeeded in making the system potential customers that the available to a much broader product is supported, that future audience than was previously versions will continue to be devel­ possible. oped, and that these versions will The freezing of what previously be upwardly compatible with all had been an ever-changing UNIX past applications. interface represented a major de­ AT&T's size alone ensures that parture from the pattern that the it will be around in years to come. small but highly skilled UNIX The fact that the company has community had come to expect. allocated a growing number of Most early users had accounts at research, development, and sup­ sites that had the source to the port resources to UNIX over the programs they ran. Thus, as the past 10 years provides an assur­ system interface evolved to reflect ance of its commitment. Mean­ more current technology, soft­ while, its massive advertising ware could be changed to keep campaign for System V, its pres­ pace. Users simply updated their ence on the /usr/group UNIX programs to account for the new standards committee, and the interface, recompiled them, and publication of the System V In­ continued to use them as before. terface Definition testify to the Although this required a large company's intention to remain effort, it allowed the system— compatible with past systems. and the tools that ran on it— Although repeal of the law of to reflect changes in software entropy is a necessary step along technology. the road to a viable commercial At the forefront of the techno­ product, this runs counter to logical wave was AT&T's own Bell

28 UNIX REVIEW OCTOBER 1985 Illustration by M. Kathryn Thompson :**£&« v* v-

^^fe^|pjp^^^^v^.'-- •••: £*•*. ; ^ ;*** IS^MIllt^-^

.->.^«i .-*-: •

!f'$&:&'l~- -'-^ 5S^*-4 St UNIVERSITY ROLE

Laboratories. It was there that was beginning to put the brakes In addition, they were able to the UNIX system was born and on further change in UNIX, local discuss current problems and nurtured, and it was there that area networks and bitmapped hash out potential solutions its evolution was controlled—up workstations were just beginning at semi-annual technical confer­ through the release of the 7th to emerge from Xerox PARC and ences run by the Usenix or­ Edition. Universities also were other research centers. Users in ganization. involved with the system almost the academic and research com­ The assistance of a steering from its inception. The University munity realized that there were committee composed of academ­ of California at Berkeley was one no production-quality operating ics, commercial vendors, DARPA of the first participants, playing systems capable of making use of researchers, and people from the host to several researchers on such hardware. They also saw Labs made it possible for the sabbatical from the Labs. This that networking unquestionably architecture of a networking- cooperation typified the harmony would be an indispensable facili­ based UNIX system to be devel­ that was characteristic of the ty in future systems research. oped. By keeping with the UNIX early UNIX community. Work that Though it was not clear that UNIX tradition of integrating work done was contributed to the Labs was the correct base on which to by others in preference to writing by different members of the build a networked system, it was everything from scratch, 4.2BSD community helped produce a rap­ clear that UNIX offered the most was released less than two years idly expanding set of tools and expedient means by which to later. facilities. build such a system. With the release of the 7th This posed the Berkeley group MECHANISMS FOR Edition, though, the usefulness of with an interesting challenge: PRODUCING ORDERLY UNIX already had been clearly how to meet the needs of the EVOLUTION established, and other organiza­ community of users without add­ Software systems have often tions within AT&T began to han­ ing needless complexity to exist­ been compared to biological or­ dle the public releases of the ing applications. Their efforts ganisms. They are born and go system. These groups took far were aided by the presence of a through a period of innocence less input from the community as akin to childhood. They then go they began to freeze the system through another burst of growth interface in preparation for entry that takes them into the adult into the commercial marketplace. Although repeal of the world where they are expected to As the research community give up their childish ways. As continued to modify the UNIX law of entropy is a people come to rely on these system, it found that it needed systems, crashes and the loss of an organization that could pro­ necessary step along data cease to be considered ac­ duce releases. Berkeley quickly ceptable behavior. As the soft­ stepped into the role. Prior to the the road to a viable ware grows into middle age, final public release of UNIX from it gains a wider exposure that the Labs, Berkeley's work had commercial product, allows it to be used in more criti­ been focused on the development cal and demanding applications. of tools designed to be added to this runs counter to During this period, a system existing UNIX systems. After the reaches the most productive part AT&T freeze, though, a group of orderly system of its life. As it ages, though, it researchers at the university becomes less able to adapt to found that they could easily ex­ evolution. changing times. Eventually, it pand their role to include the must retire so that younger, more coalescing function previously agile systems can move in to take provided by the Labs. Out of this its place. came the first full Berkeley distri­ large and diverse local group of UNIX was born in a lean and bution of UNIX (3.0BSD), com­ users who were teaching intro­ mean era; it was designed for plete with virtual memory—a ductory programming, typeset­ processors that ran at fractional first for UNIX users. The idea was ting documents, developing soft­ MIPS, with memories smaller so successful that System V even­ ware systems, and trying to build than 65 kilobytes, and 10-char- tually adopted it six years later. huge Lisp-based systems capable acter-per-second printing termi­ At the same time that AT&T of solving differential equations. nals that made interaction ago-

30 UNIX REVIEW OCTOBER 1985 nizingly slow. Given such a UNIX system that evolved far a central clearinghouse, in spite starting point, it is a tribute to the beyond its humble beginnings. of the reputation that clearing­ designers of UNIX that the system Note, though, that the path of house had for selective technical can now be found running on evolution is littered with broken scrutiny. multi-MIP processors, with mega­ carcasses. While gene mutation is Here one must distinguish bytes of memory, and multiwin- critical to the advance of the between the selection process dow bitmapped displays. There species, only one in 100 produces provided by research and com­ are several reasons why UNIX a useful feature; the rest result mercial organizations. Research has managed to stretch its bio­ in needless or detrimental organizations can base pruning logical limits to this degree. changes. The mere existence of considerations strictly on the co­ The single most important an environment for mutation is herence of a system. They need structural reason is that UNIX not enough—some organization not concern themselves with how was not written in assembly lan­ must bear responsibility for bru­ changes might affect past var­ guage. Equally important is the tally pruning the weak or useless iants of the system. Commercial fact that it was not written in a ideas. Here again UNIX was organizations, though, must en­ complex high-level language that unique. Unlike other projects be­ sure that changes will not affect could be compiled only on a large set by competing groups jealously programs built to tie in with an computer system. UNIX has suc­ guarding their work from one old interface. Thus, paging might ceeded largely because the C another, UNIX thrived in an open be a great idea, but it could cause language itself was just high-level and cooperative community will­ problems for old software that enough to allow it to be easily ing to channel its ideas through depends on the execution predic- compiled for a wide range of computer hardware, without be­ ing so complex or restrictive that systems programmers had to re­ WHEN SERIOUS PROGRAMMING vert to assembly language to get reasonable efficiency or function­ IS YOUR BUSINESS... ality. Although the success of UNIX does not stem solely from The Concurrent Euclid language for systems programming provides the fact that it was written in a the best in efficiency, portability, high-level language, the use of C reliability, and maintainability was a critical first step. Compilers running on UNIX/VAX, The second decision essential UNIX/11, VMS/VAX, with code to the extended evolution of UNIX generated for MC68000, resulted in the system's early MC6809, NS32000, 8086/8088 release from Bell Labs to other PDP-11, and soon running research environments in source on IBM-PC form. By providing source, the system's founders ensured that CONCURRENT EUCLID other organizations would not Compiler: CSRI Distribution Mgr. only be able to use the system, but Sandford Fleming Bldg 2002 also tinker with its inner work­ 10 King's College Road ings. The ease with which new Toronto, Canada M5S 1A4 ideas could be adapted into UNIX Tel: (416) 978-6985 always has been key to the changes that have been made to it. Whenever a new system would Book: come along that tried to upstage CONCURRENT EUCLID, UNIX, someone would dissect the THE UNIX SYSTEM AND TUNIS newcomer and clone its central Available from: ideas into UNIX. The unique abili­ Addison-Wesley Publishing CONCURRENT ty to use a small, comprehensible Company, Reading, MA. 01867 L I _Dj system, written in a high-level Tel: (617) 944-3700 fT u c language, in an environment swimming in new ideas led to a Circle No. 23 on Inquiry Card UNIX REVIEW OCTOBER 1985 31 ^UNIVERSITY ROLE

tability of a swap-based system, However, users have already dis­ how the output will appear. making it impossible for paging to covered that isolated PCs are far Similarly, the trend in the pro­ replace swapping; as a result, the less useful than PCs networked gram development arena is to­ complexity of supporting both together in a way allowing for ward object-based environments schemes must be maintained. As remote logins, file transfer, distri­ in which the system can main­ the system becomes more com­ bution of a pipeline across multi­ tain object dependencies, thus plex, its evolutionary paths will ple machines, and other distrib­ keeping the project's binaries, become increasingly restricted. uted computing applications. The libraries, and documentation up This is not surprising. No soft­ research version of UNIX demon­ to date as program development ware system can last forever; strated in 1982 that UNIX could proceeds. revolution is as necessary in the Another major research topic software world as storms are in is how to build systems using the physical world. The old guard several tightly coupled proces­ must eventually give way to new sors. Ideally, processors can be blood. The fact that UNIX was UNIX thrived in an used together to provide the illu­ provided with a less restrictive sion of a single larger machine. growth path in the research open and cooperative True multiprocessor support re­ environment during its critical quires that the UNIX kernel run adolescent period has probably community willing to simultaneously on all processors. doubled its life expectancy. Ulti­ The system depends on only one mately, the commercial system channel its ideas kernel being in operation, syn­ will have to get beyond its slow- chronizing through a large global printing terminal orientation and through a central memory and the selective block­ adopt the new technology, lest it ing of interrupts. This synchroni­ be superseded by an onslaught of clearinghouse. zation structure does not lend systems capable of supporting itself to running the kernel simul­ bitmapped displays. taneously on more than one Evolutionary restrictions are processor. A message-based syn­ such that the facilities of the evolve to accommodate network­ chronization structure lends It­ commercial system lag anywhere ing. But the commercial version self much more readily to a kernel from five years to an infinite continues to offer little more in running in a distributed environ­ amount of time behind the re­ the way of network functionality ment. Though several groups search systems. In an effort to than file transfer and batch-style have modified the existing kernel provide more modern facilities, remote execution. This probably to run in a multiprocessor envi­ many manufacturers have start­ will be remedied, despite the ronment, the added complexity ed marketing the 4.2BSD re­ commercial constraints, within makes further evolution difficult. search system in order to sell the next couple of years. The challenge facing UNIX re­ into more sophisticated technical The current trend in systems searchers today is the need to markets. Over the long term, it is research is to provide a variety of add functionality capable of reasonable to expect that the environments. One example is supporting environments while most useful functionality of the the text processing environment they produce kernel modifica­ research systems will be grafted offered by Apple's Macintosh. tions that will allow UNIX to into the commercial version, The UNIX user must learn an run in a multiprocessor environ­ while the research version will be extensive toolset used in conjunc­ ment. Since system complexity extended to provide as much of tion with an unforgiving shell increases much faster than the the commercial version's inter­ before diving into document pro­ size of its code, the current struc­ face as possible. duction. The user brave enough ture requiring a monolithic kernel to tackle this task needs first to may become untenable. However, THE FUTURE OF UNIX learn how to visualize the output growing familiarity with the UNIX UNIX is currently in its middle that the system's baroque type­ interface argues strongly in favor age. The commercial version of setting language will produce. By of maintaining it. UNIX has been widely adopted contrast, the Macintosh provides Some researchers now believe because it provides better func­ a menu-driven toolset interface that the revolution will come from tionality than any other PC oper­ and a text-processing interface below without end users being ating system on the market. that allow the user to see exactly any the wiser. They think that a

32 UNIX REVIEW OCTOBER 1985 new system will be developed that based synchronization requires consists of a small kernel-kernel more CPU cycles than the use of Dr. McKusick is involved in the providing only the lowest level semaphores in a shared memory. development of Berkeley UNIX as a of message passing, hardware As a result, such systems do not Research Computer Scientist for the scheduling, and virtual memory compete well in shops where Computer Systems Research Group management. The current UNIX timesharing computers are used at the University of California. kernel will be broken into several by so many people that nearly all While a graduate student, he imple­ server processes on top of the the system's services must be mented the fast file system distrib­ kernel-kernel. Sample services active on the same machine. uted on 4.2BSD and worked on the from existing UNIX systems in­ However, users are quickly Berkeley Pascal system. • clude a file system server, a moving toward having personal TCP/IP network server, and a networked workstations in which The views of the future are gleaned from terminal line server. New services CPU cycles are cheap but disk discussions with Bill Joy and Rick Rashid. might include an Ada develop­ and memory are expensive. This Acknowledgements to Charles Darwin and ment server or a document prep­ Thomas Jefferson are also in order. I thank will provide a setting in which the those people whose accomplishments, though aration environment server. Us­ performance problems of a mes­ unmentioned. contributed much to the sys­ ers familiar with existing UNIX sage-based system become unno- tem's vitality. For the last six years. I have been supported by the National Science systems could continue to run on ticeable and the benefits of rapid Foundation under grant MCS80-05144. and top of the UNIX file system server, prototyping and new integrated the Defense Advance Research Projects Agen­ while users wishing to work with­ facilities usher in a new era of cy (DoD) under ARPA Order No. 4031 moni­ tored by Naval Electronic System Command in an Ada environment could growth. under Contract No. N00039-82-C-0235. work with an object-based file system through an object-orient­ ed Ada database server. Synchro­ nization between the servers could be offered by the message interface provided by the kernel- kernel. By restructuring synchro­ nization through messages rather than by using a global-shared memory, servers could be decou­ pled from the particular proces­ sor they run on. The servers would be able to interact equally UNIX* COMMUNICATIONS well whether they were on the same processor, two processors X.25 • HASP • SNA3270 • SNA3770 running together, or on two pro­ Drop-in communication systems for MULTIBUS * based cessors separated by miles of computers. Offload the CPU intensive process of com­ network cable. munication with the HORIZON*™ Series of boards from This scheme would reduce the MORNING STAR. Complete systems include your choice of complexity of the system to a hardware and software combinations to custom fit your manageable level. Rather than data communication needs. Available for: Sun having to deal with a single Microsystems, Masscomp, Pyramid, Heurikon, Plexus, mountain of interrelated code, NCR Tower, Sperry 5000, Celerity and more. each server could be treated as Call today for more information a smaller independent module. This will mean that users will no longer be required to run a huge kernel providing many features Morning Star Technologies! Inc. they neither need nor want— 1760 Zollinger Road, Columbus, Ohio 43221 meaning that the system will consume fewer of the resources In Ohio call [614] 451-1883 TWX - BIO - 600 - 3S72 they do want. Systems such as this will not • UNIX is a Trademark of AT&T Bell Labs • MULTIBUS

UNIX REVIEW OCTOBER 1985 33 MONTHLY REPORT [ BSD'S BIG ADVENTURE: THE BAD BERKELEY-TO-BOSTON CONNECTION BY DAVID CHANDLER

he cover of last April's UNIX researchers at Berkeley took issue REVIEW depicted a garden with certain of its elements. In fact, T labyrinth of tall green hedges they thought so little of BBN's effort labeled "The Networking Maze". Our that they took the liberty of making intention was to illustrate the significant modifications (much as confusion many UNIX users have they already had done with AT&T's experienced when grappling with UNIX code). Thus was born a networking. A similar device would second version of TCP/IP. serve well in conjunction with the It was at this point, as Holmes tale of bureaucratic intrigue that would say, that the plot thickened. follows. This, too, is a story of BBN was not, if you will, pleased networking, but the implications are that its version had been altered for much broader—affecting the very purposes other than integration. existence of the Berkeley Software Berkeley researchers, of course, Distribution of UNIX. believed that their efforts had yielded in this particular labyrinth, the both Berkeley and BBN under a superior version. Meanwhile, principal players are the Defense contract to develop a networking DARPA, the distressed Advanced Research Projects scheme for a single operating matchmaker/general contractor, Agency (DARPA) and two of the system adaptable to the needs of all attempted to mediate. major beneficiaries of its budgetary DARPA contractors. Despite heavy lobbying from bounty—Bolt Baranek and Newman Under the terms of the contract, BBN, the Berkeley version of TCP/IP (BBN, the Cambridge, MA-based BBN was to write the Transmission became a de facto standard when it communications software house) Control Protocol/Internet Protocol was included as part of the official and the Computer Systems (TCP/IP) for the DoD's ARPANET 4.2BSD release in September, 1982. Research Group at the University of standard, and Berkeley was to Prior to this watershed event, the California at Berkeley. The three develop sockets and network differences between the two versions have generated a fair amount of interfaces. This delineation of duties had been significant but not drama over the course of an uneasy was in accordance with the fundamental. This soon was to five-year relationship that has International Standards Organization change. Though DARPA continued spawned the development of 4.2 and Reference Model for Open Systems to encourage Berkeley to use the 4.3BSD. During recent weeks, the Interconnection (ISO-OSI), the BBN code, splits between the two relationship nearly foundered as seven-layer specification for network implementations became tensions between Berkeley and BBN design. One could apply the analogy increasingly evident. staffers built up to the snapping of a sandwich: BBN's work on TCP/ Colonel Robert Baker, R&D point. IP was comparable to the filling— I rrrogra m Manager for DARPA's Christmas came early, though, as the network and transport layers Distributed Computing Program— DARPA announced a truce on (OSI layers three and four); and which includes the Berkeley UNIX December 18 that should maintain Berkeley's work on sockets and networking project—recalled what the relationship for a while longer— interfaces was the bread—the happened next: "There were some long enough, at least, to produce the surrounding physical, data link, things oriented toward local network much-anticipated release of 4.3BSD session, and presentation layers. performance that were put into the sometime in February. In keeping with DARPA's bidding, Berkeley version, and some oriented The details of how things went BBN developed the original toward internetwork performance sour would do justice by Franz implementation of TCP/IP, and that were put into the BBN version. Kafka, but to sort them out requires provided it to Berkeley. From that BBN integrated its version of the that we go back five years to the time forward, the two organizations code with 4.2 such that it was time before the Berkeley-to-Boston were supposed to work together to possible for someone who had 4.2 to bureaucratic labyrinth had been integrate the protocol into the whole replace the Berkeley networking erected —back to the time when of BSD's networking code. Upon code in their kernel with the code DARPA had just decided to bring reviewing the BBN code, though, from BBN. ..."

8 UNIX REVIEW THE MONTHLY REPORT

As Berkeley UNIX has evolved people have gotten the distribution three years, and—despite DARPA's over the past two or so years toward from Berkeley and then have assertions—was considered by the 4.3 release, both versions of the replaced the networking code with many to be superior. More crucial yet TCP/IP code have remained the code from BBN because it was the fact that patience on all available. DARPA, though, reached offered more functionality in the area sides was wearing thin, and the of internetwork robustness. Still, we Berkeley researchers responsible for decided that we didn't want to BSD itself had intimated that they This is a story of networking, continue supporting both versions." would not tolerate the bureaucratic At last June's Usenix meeting in sojourn much longer. but the implications affect the Portland, Berkeley researchers A compromise of sorts was announced that 4.3 would be reached. "What we did", Baker very existence of the Berkeley released officially within two months. recalled, "was send out for beta But that was before the networking testing [last August] with the two Software Distribution of UNIX. controversy came to a head. One versions of the networking code. . . . standard TCP/IP implementation We also had some people [in the had to be chosen, and the decision UNIX community] make an the decision some time ago that it clearly was not going to be easy; independent evaluation of the two would allow only one version to be while on the one hand DARPA had versions, looking both at integrated into 4.3. As Baker paid several million dollars for the functionality and the software itself. recalled, "We decided that we had BBN code and had supported it right The conclusion was that the BBN dealt with two versions for long from the start, the Berkeley code code offers some functionality that enough. You know, the situation in had been run in academic and the Berkeley code doesn't, and that the past has been that a number of commercial use under 4.2 for almost the reverse is also true. Either

\W^: 700am 130am neidea. Tom antes eortyto run a compile Julie and Dave have the same idea. \ *" at full blast. -. . 1 1 -K- r v 1 I fj^ht

* '•; •

1 - ... -r-fc: ** 1

! tk version could be used as the base for a number of months. On "Actually the Berkeley code in its for 4.3 and still provide a December 18, though—some four current form already has performance advantage over what's months after the anticipated release incorporated a number of things now in 4.2." date of 4.3—DARPA sent a formal from the BBN code. That's Kirk McKusick, a research letter to Berkeley indicating it had happened as part of the process of computer scientist at Berkeley and decided that the 4.3 release should putting together 4.3 and running it one of the chief developers of 4.3, include Berkeley's version of TCP/ through beta testing." lobbied strongly along with his IP. The decision, DARPA said, had A member of the BBN colleagues for the Berkeley version been based both on technical and development team chose not to of TCP/IP. According to McKusick, political evaluations. comment on what he called "a long of 35 beta sites testing 4.3BSD, 33 "There's been a much greater and involved story". Berkeley chose to use Berkeley TCP/IP—and degree of testing and experience researchers, though, understandably one of the other two was planning to with the Berkeley code", DARPA's were pleased—not only because the switch because of repeated system Baker explained. "Of course, it's the Berkeley version of TCP/IP had crashes with the BBN code. Included code that's been included in the been selected, but because the among the 4.3 beta test sites past distributions. So, on the basis matter finally had been resolved. running Berkeley's TCP/IP were a of the evaluations we have done, we McKusick said the major remaining number of major commercial decided to use the Berkeley code as hurdle for 4.3 relates to vendors who already had begun to the base for the networking code documentation. In the aftermath of base products on it, and who had distributed in 4.3, and to incorporate DARPA's decision, it will be expressed an interest in releasing into it whatever additional necessary to comb through some those packages. functionality we find we need from 10,000 pages to ensure that all The DARPA evaluation went on the BBN code. . . . TCP/IP references conform. Barring

.*"*.*:.- •*•:'; ^if^:-'^ ^.re users rush totog on. MONTHLY REPORT

any unforeseen complications, 4.3 is enhancements—job control, auto- why it was slow. That told us what expected to be released officially in configuration, and all that—but it we needed to tweak." February. didn't run that fast. It ended up So what is it that we can expect spending about 20 percent of its of commercial systems bearing a New Features time doing context switching, "4.3" label? The best source of whereas VMS, for example, was publicly-available details is a paper onsidering what is meant by spending about 4 percent. . . . But written by McKusick and Mike Karels the term "new", let's then 4.1 came out and offered of UC Berkeley, and Sam Leffler of C distinguish between those snappier performance. Lucasfilm (late of UCB) for the things that have been revised from "Next was 4.2. We were trying to Summer '85 Usenix Conference. A those that have been created from put in many new things and we copy is contained in the scratch. The forthcoming 4.3BSD fits wanted to do it with the minimal conference's proceedings. Those under the heading "revised". amount of code. Admittedly, 4.2 ran who seek technical details are 4.4BSD, however, likely will contain about 20 percent slower than 4.1. . . hereby referred to that paper, since things that never before have been Many of the internal structures of the we have space enough here only to released. kernel were used much more heavily discuss the general distinctions, of In commenting on the general in 4.2 than in 4.1, so a facility that which there are three. practice of generating new releases, was adequate in 4.1 became less Performance Improvements. This McKusick explained that, "We than adequate in 4.2." actually can be broken down into basically tend to alternate between With 4.3, "we started to fix the the two sub-classes of performance making major functional changes things that had turned up. We spent optimization for the general and tune-ups." most of our time analyzing the timesharing environment—changes "4.0 came out with several system and figuring out where and to the kernel and changes to the

930am K>30am A meeting ends and another Bob hits RETURN and cleans out twenty people run for their his wallet while he's waiting for terminats.They should hcwe walked. something to happen. system libraries and utilities. The libraries and utilities. On the kernel not wish to encourage rampant changes to the kernel involve— level, as McKusick et al. stated in the system cracking", we prefer not to among other things—name caching, Usenix paper, "A significant effort make the details public here. intelligent auto-siloing, process table went into improving the networking management, scheduling, clock part of the kernel. The work Insights with Foresight handling, file system management, consisted of fixing bugs, tuning the networking, the exec command, algorithms, and revamping the lowest arter George mentions in context switching, setjmp and levels of the system to better handle this month's feature article longjmp calls, and various heterogeneous network topologies." C on communications compensations for the lagging With regard to extensions to libraries futuristics that one does well when compiler technology of the system. and utilities, the developers wrote, discussing the future to avoid Improvements to libraries and "Most of the changes to the utilities sounding like Jeanne Dixon. There utilities, meanwhile, have resulted and libraries have been made to are, however, certain trends in from work done on hashed allow them to handle a more general computing that lend themselves to databases, buffered I/O, mail set of problems, or to handle the extrapolation, and Berkeley UNIX system software, network servers, same set of problems more quickly." not only relates to many of these the C runtime library, and the C Tightened Security. Certain trends, but often helps to shape shell. changes also were made to both the them. BSD UNIX developer Functional Extensions. As with kernel and the system's utilities in McKusick holds a seasoned the performance improvements order to enhance security. These perspective on what may lie on the made for the system, functional changes have been discussed in other side of the horizon, and is in a extensions have been divided into modest detail elsewhere, but as the position to influence decisions at extensions to a) the kernel and b) Usenix paper noted, "Since we do Berkeley. Hence, we asked him to

HOOam Witti response time this slow, TOO MANY USERS everyone may as wen goto lunch. CAN SLOW THINGS DOWN TOATRICKLE Most UNDHwsed computers Pyramid 98X And ifs a fraction can really pour it on. When you're of the cost of getting everybody the only onetogged in. a workstation. But as soon as the rest of the Naturally, it supports UNIX 4.2, gang shows up, all mat perform­ System V, PC networking, and a ance goes right ckjwn me drain. flood of other options. Best of alt Theirs when you need The you can add processors quickly, Practical ParaUeT from Sequent. inexpensively, and easily as your ., Because it's a true parallel com­ needs grow. puter, The Practical Parallel gives If you don't want bad response you more consistent response time to hang you out to dry, call man any single-processor sys­ us at 800/854-0428. Or write tem At 8 MIPs, It has enough to Sequent 15450 S.W.Koll performance to keep over a Parkway, Beaverton, Oregon ' ihundred users running full blast. 97006-6063 and ask about But unlike other 8-M1P comput- The Practical Parallel. ] er$, you a»n pet airs without get- Ifs the system that never runs \ fing soaked Our systems start at hot and cold. wel under a hundred grand. Our ptee/perfbrmonce is four Times better than a W 8600 Tire Practical PoraHeT and two times better than a See Us at UniFbrum booth *1562. 1t» faciMftKtt tuiaderaafka^

Circle No. B on inquiry Card I THE MONTHLY REPORT

expound on what he sees in BSD's which, for example, AT&T's Remote physical memory is significantly future: File System does—but suffer with different than what we've done in the Remote File Systems. "There are abysmal throughput (like 20 or 30 past. three major directions in which we KB per second).... We're not "Our question basically becomes: are going", McKusick noted. "The willing to take that kind of how do you design a Virtual throughput just because it's remote, memory' system, given that you so we might be willing to whittle have lots of memory? We'll try to 'Most of the changes have been away at some of the UNIX come up with a design that semantics. It's not exactly clear yet accomplishes that. At the same time made to allow utilities and what that's going to mean, but we'll we're going to throw in various try to restrict our changes to things things that come along for free- libraries to handle a more on which we don't think a lot of things like copy-on-write, lightweight programs depend." processes, and shared memory." general set of problems." Virtual memory. "The second When asked if Berkeley would thing we're looking at is virtual include a shared memory design memory", McKusick explained. "We similar to the one included in System first is remote file systems— basically feel that the next iteration V, McKusick responded, "It something akin to Sun's NFS. Many of virtual memory needs to address a depends: if we choose to implement of these systems have been done, whole different set of concerns, it in the way AT&T has done in and there is this continuum: the which can be summarized by System V, then we can make it easy developer either can get them to run observing that memory is getting for people to pick up. If we don't, it really fast with terrible semantics, or cheap and large very quickly will be less easy. Whenever AT&T can get them semantically correct— Therefore, what we want to do with has defined a facility before we do,

A powerful/terminol- •„.>.-' independent application : deyelppme||J.systern, Reliant helps you help yourself. You develop and install new ^applications easily and in a fractionjof the time - e that would be needed without Reliant, Refibnt also makes ^?:^ you Je?s dependent oh ^pensive application programmers. ;

Reliant consists of four distinct but interrelated parts: a Data Definition System to rapidly define data to be entered; a General Data Entry Layer to enter revise and delete data without error; a Report Generator to quickly specify the data to be retrieved and in what form; and a Document Generator to create the form of data presentation. ^ v • v 'Tv^ Best pf all, because

yduget UPDATES FREE, 1ofJM{ ^V : C" ..J^;X\.-,:it-f2^^-, Jo try Reliant Jiv your own^ironm^nt; Send us a $100.00 handling fee. VVeli rush you the latest^driubls and software to try for 30 days! Xyou agfe^Jlt^dlly is the $)swer jo.^ yv£ll credit the $100.00.to ifi^purchase pride.

Qfjtourse,th!$ best of software is* worthlessJ you dohl have the

f I •ii 91 •iwm 11^ fi» A iX» If »; 9JICl*Anid9Iiim*\9]k A~f+WtltZim9]i9]9iL~iiiH 3^!'M'^^ifyli^S9^Pu^^^^ the best power/performance raMa\the ^ cV^^" : ,,-\j : 7^$ from a variety of components fb v::'^f^ZV :-'X^ \ **i$j$tomcohfigure your system, LI k . w .' .•;'. S^r L^VV, \u\^'^i^i^'--:,;'- X^ibfind piit to*'^ / ;U^x ^ registeredfrdafroark qf;AT&f ^i Laboratories ,r:^^^ ^£J5^ystr(^^n(t.\ft|1l s|rfej jlplpiir.'foiled cof9l6g.-_• -•".•--'"v-^/.;:!."-..-Miigof.%"curreYitly

Circle No. 9 on Inquiry Card

14 UNIX REVIEW we have tried to implement a this connecting ability but says matter. compatible interface or a broader nothing about what's inside. AT&T The next BSD release, then, may set of primitives from which the —as far as I can tell— [has it well develop just in the way that its interface-could be built. We deviate structured so that] the pieces will go 4.2 forebear did. It probably will from this policy only when we feel together just like railroad cars—you offer major functional enhancements that there is a significant technical can connect them together, but what over its immediate predecessor, but advantage to an incompatible you load inside can differ from only at the expense of performance. design." boxcar to boxcar. The things that we McKusick acknowledged that, "In Stackable Line Disciplines. This is will provide inside the boxcars will be 4.3, we strove to provide binary a third likely area for future Berkeley based on what we currently have in compatibility with 4.2, so that one UNIX development. "System V calls BSD, while AT&T will load in things would be able to take a 4.2 binary it streams", McKusick explained. based on its view of how networking and run it easily on 4.3. But in 4.4, Does this mean that Berkeley is should be. AT&T's base primitives one will—at a minimum—need to planning to adopt AT&T's streams for networking are virtual circuits, recompile certain programs. It may technology? Conceptually, perhaps, whereas ours are datagrams." even be necessary to make textual but literally. . .no. McKusick Would there then be compatibility changes in a small number of them. elaborated: "There are two parts to between the Berkeley and AT & T The system will offer vast new it—the mechanism and the content. disciplines? "One should be functionality, but it undoubtedly will The mechanism is the ability to plug able to pick boxes up out of System have a performance drop. That these things together; the content is V and with a very small amount of always happens." what one puts inside the things that work put them into a Berkeley are plugged together. Streams—or system, and vice versa. Whether one David Chandler is the Associate stackable line disciplines—provides would want to do that is another Editor of UNIX REVIEW.

UBACKUP UNIX BACKUP. RESTORE, AND MEDIA MANAGEMENT SYSTEMS USECURE UTILITY SYSTEM SECURITY MANAGEMENT SPR SOFTWARE PRINT SPOOLING AND BATCH JOB SCHEDULING -SO YOU SSL FULL-SCREEN APPLICATION DEVELOPMENT

CAN GET S-TELEX ON WITH TELEX COMMUNICATIONS MANAGEMENT SSE YOUR JOB. FULL-SCREEN TEXT EDITOR For more information, These products are available for most UNIX or UNIX-derivative operating systems, including System V. 4.2 BSD. 4.1 BSD, Xenix. Version 7. System III call or write. Uniplus. and others. (703)734-9844 UNIX is a trademark of AT&T Bell Laboratories LNITECH Visit us at UniForum Booth No. 1151 8 33 0 OLDCOURTHOUSJERD SUITE800 VIENNA. VIRGINIA

Circle No. 10 on Inquiry Card

JANUARY 1986 15