DOCSLIB.ORG

Energy Efficient Cloud Control and Pricing in Geographically

Total Page：16

File Type：pdf, Size：1020Kb

Download full-text PDF Read full-text
Energy Efficient Cloud Control and Pricing in Geographically Energy Efficient Cloud Control and Pricing in Geographically Distributed Data Centers DISSERTATION zur Erlangung des akademischen Grades Doktor der Technischen Wissenschaften eingereicht von Dražen Lucaninˇ Matrikelnummer 0848810 an der Fakultät für Informatik der Technischen Universität Wien Betreuung: Univ. Prof. Dr. Ivona Brandic´ arXiv:1809.05853v1 [cs.DC] 16 Sep 2018 Diese Dissertation haben begutachtet: Univ. Prof. Dr. Ivona Brandic´ Univ. Prof. Dr. Helmut Hlavacs Technische Universität Wien A-1040 Wien Karlsplatz 13 Tel. +43-1-58801-0 www.tuwien.ac.at Wien, 21. April 2016 Dražen Lucaninˇ Energy Efficient Cloud Control and Pricing in Geographically Distributed Data Centers DISSERTATION submitted in partial fulfillment of the requirements for the degree of Doktor der Technischen Wissenschaften by Dražen Lucaninˇ Registration Number 0848810 to the Faculty of Informatics at the Vienna University of Technology Advisor: Univ. Prof. Dr. Ivona Brandic´ The dissertation has been reviewed by: Univ. Prof. Dr. Ivona Brandic´ Univ. Prof. Dr. Helmut Hlavacs Technische Universität Wien A-1040 Wien Karlsplatz 13 Tel. +43-1-58801-0 www.tuwien.ac.at Vienna, 21st April, 2016 Dražen Lucaninˇ Erklärung zur Verfassung der Arbeit Dražen Lucaninˇ Bürgerspitalgasse 22/15, 1060 Wien Hiermit erkläre ich, dass ich diese Arbeit selbständig verfasst habe, dass ich die verwendeten Quellen und Hilfsmittel vollständig angegeben habe und dass ich die Stellen der Arbeit – einschließlich Tabellen, Karten und Abbildungen –, die anderen Werken oder dem Internet im Wortlaut oder dem Sinn nach entnommen sind, auf jeden Fall unter Angabe der Quelle als Entlehnung kenntlich gemacht habe. Wien, 21. April 2016 Dražen Lucaninˇ i Acknowledgements The research leading to this thesis has received funding from the following projects: (1) the “Holistic Energy Efficient Approach for the Management of Hybrid Clouds” (HALEY) project, granted under the Vienna University of Technology research award; (2) several international meeting and training school grants by COST Action IC804 “Energy efficiency in large scale distributed systems”; (3) the short term scientific mission (STSM) grant by COST Action IC1305 “Network for Sustainable Ultrascale Computing” (NESUS). I would like to thank my supervisor Univ. Prof. Dr. Ivona Brandic´ for her support and mentoring during my research at the Vienna University of Technology. I would also like to thank Univ. Prof. Dr. Helmut Hlavacs for providing feedback as an external reviewer of this thesis. Additionally, I would like to thank Senior Lecturer Dr. Rizos Sakellariou for our research collaboration and for hosting my research visit at the University of Manchester. I am also grateful to my research collaborators Ilia Pietri, Simon Holmbacka and Foued Jrad for their help in our joint work. Big thanks go to my close colleagues and collaborators Soodeh Farkohi, Toni Mastelic,´ Ivan Breškovic,´ Michael Maurer and Vincent Chimaobi Emeakaroha for the discussions, talks and fun we had during our period of working together. Additionally, I would like to thank the other members of the research groups I have had the privilege of collaborating with – the Distributed Systems Group and the Electronic Commerce Group at the Vienna University of Technology and the Division of Electronics at the Ruder¯ Boškovic´ Institute in Zagreb, Croatia. Finally, I would like to thank my partner Sara, my family and my friends who were always there for me when I needed advice or encouragement. I also have to thank my favorite musicians and bands who created the many great songs that were a musical inspiration during my work. Dražen Lucaninˇ April 2016 iii Abstract The fast pace of progress in the domain of applications provided over the Internet has created a need for computational resources delivered as an on-demand utility, without having to manually manage computer hardware. Cloud computing has emerged as a very popular paradigm where resources such as virtual machines are provided as a scalable, pay-as-you-go service, catering to applications in a multitude of fields. On the other hand, the rapid cloud computing growth has turned the energy consumption of data centers hosting the cloud’s hardware infrastructure into a global environmental problem and a major cost factor. It is estimated that data centers constitute 1.5% of global electricity usage. At the same time, to serve increasing user requirements, modern cloud providers are operating multiple geographically distributed data centers. Distributed data center infrastructure changes the rules of cloud control, as energy costs depend on current regional electricity prices and temperatures that we call geotemporal inputs. Furthermore, pricing policies at which cloud providers can offer computational resources depend on the quality of service (QoS). With such pricing schemes and the increasing energy costs in data centres, balancing energy savings with performance and revenue losses is a challenging problem for cloud providers. Existing cloud control methods are suitable only for a single data center or do not consider all the available cloud control actions that can reduce energy costs in geographically distributed data centers. In this thesis, we propose a pervasive cloud control approach consisting of multiple methods for dynamic resource reallocation and hardware configuration adapted to volatile geotemporal inputs. The proposed methods consider the QoS impact of cloud control actions and the data quality limits of time series forecasting methods. We offer a cloud controller design that supports future extensions when new decision support components need to be added. We also propose novel pricing schemes which account for the computational resource availability and costs that arise from our cloud control approach to enable both flexible, energy-aware and high performance cloud computing. We evaluate our cloud control methods empirically and in a number of simulations using historical traces of electricity prices, temperatures, workloads and other data. We estimate the potential energy cost savings by comparing our methods to state-of-the-art baseline methods. We explore a variety of input parameters to provide a range of guidelines for practical application of our methods in cloud systems. Our results show that significant energy cost savings are possible without harming the QoS or service revenue in geographically distributed cloud computing. v Kurzfassung Der schnelle Fortschritt im Bereich von Internet-Anwendungen hat es erforderlich gemacht, Re- chenressourcen on-demand zur Verfügung zu stellen ohne Computer-Hardware manuell managen zu müssen. Cloud Computing hat sich als sehr beliebtes Paradigma herauskristallisiert, um Ressourcen wie virtuelle Maschinen mannigfaltigen Anwendungen skalierbar in Form eines "Pay-as-you-goSer- vices zugänglich zu machen. Andererseits hat das rapide Wachstum von Cloud Computing den Energieverbrauch von Da- tenzentren, die die Cloud-Hardware betreiben, zu einem globalen Umweltproblem und einem gewichtigen Kostenfaktor gemacht. Man schätzt, dass Datenzentren 1,5% des globalen Elektrizi- tätsverbrauchs ausmachen. Gleichzeitig betreiben moderne Cloud-Provider mehrere geographisch verteilte Datenzentren, um steigenden Nutzeranforderungen gerecht zu werden. Eine verteilte Infrastruktur von Datenzentren verändert die Cloud-Kontrollregeln, da Energiekosten von den aktuellen regionalen Elektrizitätspreisen und Temperaturen (sog. geozeitliche Einflussfaktoren) abhängen. Zusätzlich hängt die Preispolitik, über die Cloud-Provider Rechenressourcen anbieten, vom Quality-of-Service (QoS) ab. Auf Grund dieser Preisschemata und der steigenden Energiekos- ten in Datenzentren, ist das Austarieren zwischen Energieersparnissen und damit einhergehenden Verlusten an Leistung und Gewinnen ein herausforderndes Problem für Cloud-Provider. Bestehende Cloud-Kontrollmethoden existieren nur für einzelne Datenzentren oder berücksichtigen die ver- fügbaren Cloud-Kontrollaktionen nicht, die zur Reduzierung von Energiekosten in geographisch verteilten Datenzentren führen können. In dieser Dissertation schlagen wir eine durchdringende Cloud-Kontrollmethode vor, die aus mehreren Methoden für dynamische Resourcereallokation und Hardwarekonfiguration basierend auf volatilen geozeitlichen Einflussfaktoren besteht. Die vorgestellten Methoden betrachten den QoS- Einfluss von Cloud-Kontrollaktionen, sowie Limits für die Datenqualität von Vorhersagemethoden von Zeitreihen. Wir präsentieren einen Cloud-Kontrollmechanismus, der zukünftige Erweiterungen, falls neue Komponenten zur Entscheidungsfindung hinzugefügt werden müssen, ermöglicht. Außer- dem stellen wir neuartige Preisschemata vor, die die Verfügbarkeit der Rechenressourcen, sowie die Kosten, die durch unsere Cloud-Kontrollmethode entstehen, in Betracht ziehen, um flexibles, energiebewusstes und hochperformantes Cloud-Computing zu ermöglichen. Die Cloud-Kontrollmethoden werden empirisch und durch Simulationen evaluiert, die u.a. auf historischen Daten von Elektrizitätspreisen, Temperaturen und Auslastung beruhen. Die potentiellen Energiekostenersparnisse werden geschätzt, indem unsere Methoden mit Methoden, die Stand der vii Technik sind, verglichen werden. Wir untersuchen eine Vielzahl an Einflussparametern, um eine Reihe von Empfehlungen für die praktische Verwendung unserer Methoden in Cloud-Systemen abzu- geben. Die Resultate zeigen, dass signifikante Energiekostenersparnisse möglich sind, ohne das QoS oder die Gewinne von Services im geographisch verteilten Cloud-Computing
Load more

Recommended publications
  • Raising the Bar for Using Gpus in Software Packet Processing Anuj Kalia and Dong Zhou, Carnegie Mellon University; Michael Kaminsky, Intel Labs; David G
    Raising the Bar for Using GPUs in Software Packet Processing Anuj Kalia and Dong Zhou, Carnegie Mellon University; Michael Kaminsky, Intel Labs; David G. Andersen, Carnegie Mellon University https://www.usenix.org/conference/nsdi15/technical-sessions/presentation/kalia This paper is included in the Proceedings of the 12th USENIX Symposium on Networked Systems Design and Implementation (NSDI ’15). May 4–6, 2015 • Oakland, CA, USA ISBN 978-1-931971-218 Open Access to the Proceedings of the 12th USENIX Symposium on Networked Systems Design and Implementation (NSDI ’15) is sponsored by USENIX Raising the Bar for Using GPUs in Software Packet Processing Anuj Kalia, Dong Zhou, Michael Kaminsky∗, and David G. Andersen Carnegie Mellon University and ∗Intel Labs Abstract based implementations are far easier to experiment with) and in practice (software-based approaches are used for Numerous recent research efforts have explored the use low-speed applications and in cases such as forwarding of Graphics Processing Units (GPUs) as accelerators for within virtual switches [13]). software-based routing and packet handling applications, Our goal in this paper is to advance understanding of typically demonstrating throughput several times higher the advantages of GPU-assisted packet processors com- than using legacy code on the CPU alone. pared to CPU-only designs. In particular, noting that In this paper, we explore a new hypothesis about such several recent efforts have claimed that GPU-based de- designs: For many such applications, the benefits arise signs can be faster even for simple applications such as less from the GPU hardware itself as from the expression IPv4 forwarding [23, 43, 31, 50, 35, 30], we attempt to of the problem in a language such as CUDA or OpenCL identify the reasons for that speedup.
    [Show full text]
  • Computer Architecture: Parallel Processing Basics
    Computer Architecture: Parallel Processing Basics Onur Mutlu & Seth Copen Goldstein Carnegie Mellon University 9/9/13 Today What is Parallel Processing? Why? Kinds of Parallel Processing Multiprocessing and Multithreading Measuring success Speedup Amdhal’s Law Bottlenecks to parallelism 2 Concurrent Systems Embedded-Physical Distributed Sensor Claytronics Networks Concurrent Systems Embedded-Physical Distributed Sensor Claytronics Networks Geographically Distributed Power Internet Grid Concurrent Systems Embedded-Physical Distributed Sensor Claytronics Networks Geographically Distributed Power Internet Grid Cloud Computing EC2 Tashi PDL'09 © 2007-9 Goldstein5 Concurrent Systems Embedded-Physical Distributed Sensor Claytronics Networks Geographically Distributed Power Internet Grid Cloud Computing EC2 Tashi Parallel PDL'09 © 2007-9 Goldstein6 Concurrent Systems Physical Geographical Cloud Parallel Geophysical +++ ++ --- --- location Relative +++ +++ + - location Faults ++++ +++ ++++ -- Number of +++ +++ + - Processors + Network varies varies fixed fixed structure Network --- --- + + connectivity 7 Concurrent System Challenge: Programming The old joke: How long does it take to write a parallel program? One Graduate Student Year 8 Parallel Programming Again?? Increased demand (multicore) Increased scale (cloud) Improved compute/communicate Change in Application focus Irregular Recursive data structures PDL'09 © 2007-9 Goldstein9 Why Parallel Computers? Parallelism: Doing multiple things at a time Things: instructions,
    [Show full text]
  • DISTRIBUTED COMPUTING ENVIRONMENT ABSTRACT The
    DISTRIBUTED COMPUTING ENVIRONMENT ABSTRACT The high volume of networked computers, workstations, LANs has prompted users to move from a simple end user computing to a complex distributed computing environment. This transition is not just networking the computers, but also involves the issues of scalability, security etc. A Distributed Computing Environment herein referred to, as DCE is essentially an integration of all the services necessary to develop, support and manage a distributed computing environment. This term paper discusses the three important issues addressed by DCE in detail, Remote Procedure Calls [IRPC], Distributed File Systems [IDFS][OSF91] and Security [Isec][OSF92]. 1 INTRODUCTION The present day computing industry depends on the efficient usage of resources. So instead of duplicating the resources at every node of computing, a remote method of accessing the resources is more efficient and saves costs. This gave rise to the field of distributed computing, where not only physical resources, but also processing power was distributed. Distributed computing was driven by the following factors. a) Desire to share data and resources b) Minimize duplication of functionality c) Increase cost efficiency d) Increase reliability and availability of resources. The Open Source Foundation’s DCE (OSF DCE) has become the De facto standard for DCE applications and has the backing of IBM, COMPAQ, HP and the likes [OSFInter92]. 1.1 Why DCE? When an organization migrates from networked computing to distributed computing a lot of factors are to be taken into consideration. For example replication of files gives rise to consistency problems, clock synchronization becomes important, and security is a bigger consideration.
    [Show full text]
  • Cost Optimization Pillar AWS Well-Architected Framework
    Cost Optimization Pillar AWS Well-Architected Framework July 2020 This paper has been archived. The latest version is now available at: https://docs.aws.amazon.com/wellarchitected/latest/cost-optimization-pillar/welcome.htmlArchived Notices Customers are responsible for making their own independent assessment of the information in this document. This document: (a) is for informational purposes only, (b) represents current AWS product offerings and practices, which are subject to change without notice, and (c) does not create any commitments or assurances from AWS and its affiliates, suppliers or licensors. AWS products or services are provided “as is” without warranties, representations, or conditions of any kind, whether express or implied. The responsibilities and liabilities of AWS to its customers are controlled by AWS agreements, and this document is not part of, nor does it modify, any agreement between AWS and its customers. © 2020 Amazon Web Services, Inc. or its affiliates. All rights reserved. Archived Contents Introduction .......................................................................................................................... 1 Cost Optimization ................................................................................................................ 2 Design Principles.............................................................................................................. 2 Definition ..........................................................................................................................
    [Show full text]
  • Cost Optimization Pillar AWS Well-Architected Framework Cost Optimization Pillar AWS Well-Architected Framework
    Cost Optimization Pillar AWS Well-Architected Framework Cost Optimization Pillar AWS Well-Architected Framework Cost Optimization Pillar: AWS Well-Architected Framework Copyright © Amazon Web Services, Inc. and/or its affiliates. All rights reserved. Amazon's trademarks and trade dress may not be used in connection with any product or service that is not Amazon's, in any manner that is likely to cause confusion among customers, or in any manner that disparages or discredits Amazon. All other trademarks not owned by Amazon are the property of their respective owners, who may or may not be affiliated with, connected to, or sponsored by Amazon. Cost Optimization Pillar AWS Well-Architected Framework Table of Contents Abstract and Introduction ................................................................................................................... 1 Abstract .................................................................................................................................... 1 Introduction .............................................................................................................................. 1 Cost Optimization .............................................................................................................................. 3 Design Principles ........................................................................................................................ 3 Definition .................................................................................................................................
    [Show full text]
  • GPU Nodes for GROMACS Biomolecular Simulations
    Best bang for your buck: GPU nodes for GROMACS biomolecular simulations Carsten Kutzner,∗,† Szilárd Páll,‡ Martin Fechner,† Ansgar Esztermann,† Bert L. de Groot,† and Helmut Grubmüller† Theoretical and Computational Biophysics, Max Planck Institute for Biophysical Chemistry, Am Fassberg 11, 37077 Göttingen, Germany, and Theoretical and Computational Biophysics, KTH Royal Institute of Technology, 17121 Stockholm, Sweden E-mail: [email protected] Abstract The molecular dynamics simulation package GROMACS runs efficiently on a wide vari- ety of hardware from commodity workstations to high performance computing clusters. Hard- ware features are well exploited with a combination of SIMD, multi-threading, and MPI-based SPMD / MPMD parallelism, while GPUs can be used as accelerators to compute interactions offloaded from the CPU. Here we evaluate which hardware produces trajectories with GRO- arXiv:1507.00898v1 [cs.DC] 3 Jul 2015 MACS 4.6 or 5.0 in the most economical way. We have assembled and benchmarked com- pute nodes with various CPU / GPU combinations to identify optimal compositions in terms of raw trajectory production rate, performance-to-price ratio, energy efficiency, and several other criteria. Though hardware prices are naturally subject to trends and fluctuations, general ten- dencies are clearly visible. Adding any type of GPU significantly boosts a node’s simulation ∗To whom correspondence should be addressed †Theoretical and Computational Biophysics, Max Planck Institute for Biophysical Chemistry, Am Fassberg 11, 37077 Göttingen, Germany ‡Theoretical and Computational Biophysics, KTH Royal Institute of Technology, 17121 Stockholm, Sweden 1 performance. For inexpensive consumer-class GPUs this improvement equally reflects in the performance-to-price ratio. Although memory issues in consumer-class GPUs could pass un- noticed since these cards do not support ECC memory, unreliable GPUs can be sorted out with memory checking tools.
    [Show full text]
  • Incremental Parallelization with Migration
    UC Irvine UC Irvine Electronic Theses and Dissertations Title Incremental Parallelization with Migration Permalink https://escholarship.org/uc/item/82w6f2w0 Author Zhang, Wenhui Publication Date 2014 Peer reviewed|Thesis/dissertation eScholarship.org Powered by the California Digital Library University of California UNIVERSITY OF CALIFORNIA, IRVINE Incremental Parallelization with Migration DISSERTATION submitted in partial satisfaction of the requirements for the degree of DOCTOR OF PHILOSOPHY in Information and Computer Science by Wenhui Zhang Dissertation Committee: Professor Lubomir F. Bic, Co-Chair Professor Michael B. Dillencourt, Co-Chair Professor Amelia C. Regan 2014 Portion of Chapter 2 c 2011 IEEE Computer Society Portion of Chapter 3 c 2012 IEEE Computer Society 2005 IEEE Computer Society Portion of Chapter 4 c 2012 IEEE Computer Society 2005 IEEE Computer Society Portion of Chapter 5 c 2011 IEEE Computer Society All other materials c 2014 Wenhui Zhang DEDICATION This dissertation is dedicated to my husband, Chuanwen, who has persistently supported me and understood the challenges and sacrifices; our lovely son, who is a smart and handsome boy and who wished his mom could have spent more time with him; my father who has been always encouraged me to pursue higher and higher and my mother who has offered her love without expecting anything back. ii TABLE OF CONTENTS Page LIST OF FIGURES v LIST OF TABLES vii ACKNOWLEDGMENTS viii CURRICULUM VITAE ix ABSTRACT OF THE DISSERTATION xi 1 Introduction 1 1.1 Research Motivation and Target Problem . 1 1.2 Dissertation Overview and Organization . 4 2 Navigational Programming 5 2.1 Overview . 5 2.2 Characteristics .
    [Show full text]
  • On I/O Performance and Cost Efficiency of Cloud Orst Age: a Client's Perspective
    Louisiana State University LSU Digital Commons LSU Doctoral Dissertations Graduate School 11-4-2019 On I/O Performance and Cost Efficiency of Cloud orSt age: A Client's Perspective Binbing Hou Louisiana State University and Agricultural and Mechanical College Follow this and additional works at: https://digitalcommons.lsu.edu/gradschool_dissertations Part of the Computer and Systems Architecture Commons, and the Data Storage Systems Commons Recommended Citation Hou, Binbing, "On I/O Performance and Cost Efficiency of Cloud Storage: A Client's Perspective" (2019). LSU Doctoral Dissertations. 5090. https://digitalcommons.lsu.edu/gradschool_dissertations/5090 This Dissertation is brought to you for free and open access by the Graduate School at LSU Digital Commons. It has been accepted for inclusion in LSU Doctoral Dissertations by an authorized graduate school editor of LSU Digital Commons. For more information, please [email protected]. ON I/O PERFORMANCE AND COST EFFICIENCY OF CLOUD STORAGE: A CLIENT'S PERSPECTIVE A Dissertation Submitted to the Graduate Faculty of the Louisiana State University and Agricultural and Mechanical College in partial fulfillment of the requirements for the degree of Doctor of Philosophy in The Department of Computer Science and Engineering by Binbing Hou B.S., Wuhan University of Technology, 2011 M.S., Huazhong University of Science and Technology, 2014 December 2019 Acknowledgments At the end of the doctorate program, I would like to express my deepest gratitude to many people. It is their great help and support that make this dissertation possible. First of all, I would like to thank my advisor, Dr. Feng Chen, for his continuous support.
    [Show full text]
  • Parallelism Analysis of Prominent Desktop Applications: an 18-Year Perspective
    Parallelism Analysis of Prominent Desktop Applications: An 18-Year Perspective Siying Feng, Subhankar Pal, Yichen Yang, Ronald G. Dreslinski University of Michigan, Ann Arbor, MI, USA ffengsy, subh, yangych, [email protected] Abstract—Improvements in clock speed and exploitation of User requirements are significantly diverse. Gamers, for Instruction-Level Parallelism (ILP) hit a roadblock during mid- example, require a high-end machine with an advanced GPU 2000s. This, coupled with the demise of Dennard scaling, led and efficient cooling. On the other hand, for a user who uses to the rise of multi-core machines. Today, multi-core processors are ubiquitous and architects have moved to specialization to their system primarily to browse the web and watch videos, it work around the walls hit by single-core performance and chip is cost-inefficient to own a system with a high-end GPU. Thus, Thermal Design Power (TDP). The pressure of innovation in the it is worthwhile to analyze the characteristics of commonly aftermath of Dennard scaling is shifting to software developers, used applications tailored toward different users. who are required to write programs that make the most effective We repeat some of the experiments of an eight-year old use of underlying hardware. This work presents quantitative and qualitative analyses of how software has evolved to reap the work by Blake et al. [3], characterizing the parallelism ex- benefits of multi-core and heterogeneous computers, compared ploited by software in desktop workstations. To analyze how to state-of-the-art systems in 2000 and 2010. We study a wide a wide spectrum of commonly used desktop applications have spectrum of commonly-used applications on a state-of-the-art evolved to utilize available hardware, we use the metrics of desktop machine and analyze two important metrics, Thread- Thread Level Parallelism (TLP) and GPU utilization.
    [Show full text]
  • Efficiency Matters!
    Efficiency Matters! Eric Anderson and Joseph Tucek Hewlett-Packard Laboratories 1501 Page Mill Road, Palo Alto, CA 94304, USA ABSTRACT The gap in theoretical per node performance and what is Current data intensive scalable computing (DISC) systems, actually achieved has become glaringly large. Expending although scalable, achieve embarrassingly low rates of pro- further research effort to support such clusters (e.g., advance cessing per node. We feel that current DISC systems have energy savings, analyze their workloads, etc.) is less useful repeated a mistake of old high-performance systems: focus- if orders of magnitude of efficiency increases on the single ing on scalability without considering efficiency. This poor node are not only possible, but easy. Further, inexpensive efficiency comes with issues in reliability, energy, and cost. 12-, 16-, and 24-core dual-socket machines are around the As the gap between theoretical performance and what is ac- corner, and will exacerbate the wastefulness of current DISC tually achieved has become glaringly large, we feel there is designs. We define a variety of types of efficiency, from CPU a pressing need to rethink the design of future data inten- and IO to cost and human resources, and show that existing sive computing and carefully consider the direction of future DISC systems do not perform well according to many of research. these measures. 1. INTRODUCTION We feel there is a pressing need to rethink the design of future data intensive computers and carefully consider the Canonically instantiated by MapReduce [9], and defined in [7], 1 direction of future research because the poor efficiency of data intensive scalable computing (DISC) focuses on large existing systems is creating avoidable problems and hiding amounts of data.
    [Show full text]
  • SC20-Final-Program-V2.Pdf
    Table of Contents ACM Gordon Bell COVID Finalist Keynote ACM Gordon Bell Finalist More Than HPC Plenary ACM Student Research Competition: Panel Graduate Posters Paper ACM Student Research Competition: Research Posters Undergraduate Posters Scientific Visualization Awards Presentation & Data Analytics Showcase Birds of a Feather SCinet Booth Sessions State of the Practice Talk Doctoral Showcase Students@SC Early Career Program Test of Time Exhibitor Forum Tutorial Exhibits Virtual Student Cluster Competition Invited Talk Workshop Job Posting ACM Gordon Bell COVID Finalist (back to top) Thursday, November 19th 10:00 am - 12:00 pm Gordon Bell COVID-19 Prize Finalist Session 1 Session Description: Enabling Rapid COVID-19 Small Molecule Drug Design Through Scalable Deep Learning of Generative Models Sam Ade Jacobs (Lawrence Livermore National Laboratory), Tim Moon (Lawrence Livermore National Laboratory), Kevin McLoughlin (Lawrence Livermore National Laboratory), Derek Jones (Lawrence Livermore National Laboratory), David Hysom (Lawrence Livermore National Laboratory), Dong H. Ahn (Lawrence Livermore National Laboratory), John Gyllenhaal (Lawrence Livermore National Laboratory), Pythagoras Watson (Lawrence Livermore National Laboratory), Felice C. Lightsone (Lawrence Livermore National Laboratory), Jonathan E. Allen (Lawrence Livermore National Laboratory), Ian Karlin (Lawrence Livermore National Laboratory), Brian Van Essen (Lawrence Livermore National Laboratory) We improved the quality and reduced the time to produce machine-learned models for use in small molecule antiviral design. Our globally asynchronous multi-level parallel training approach strong scales to all of Sierra with up to 97.7% efficiency. We trained a novel, character-based Wasserstein autoencoder that produces a higher quality model trained on 1.613 billion compounds in 23 minutes while the previous state-of-the-art takes a day on 1 million compounds.
    [Show full text]
  • Cost-Aware Scalability of Applications in Public Clouds
    Cost-aware scalability of applications in public clouds Daniel Moldovan, Hong-Linh Truong, Schahram Dustdar Distributed Systems Group, TU Wien E-mail: fd.moldovan, truong, [email protected] Abstract—Scalable applications deployed in public clouds can just manipulate their number [5]. However, cost of scalable be built from a combination of custom software components applications is complex, some services having multiple cost and public cloud services. To meet performance and/or cost elements. E.g., a VM service could be billed both every hour, requirements, such applications can scale-out/in their components during run-time. When higher performance is required, new and separately per each GB of generated I/O. Certain costs can component instances can be deployed on newly allocated cloud be static, such as costs for reserving a cloud service [6], [7], services (e.g., virtual machines). When the instances are no [8]. Other costs can be dynamic, such as discounts for certain longer needed, their services can be deallocated to decrease cost. service usage levels. Costs of cloud services can also depend However, public cloud services are usually billed over pre-defined on service combinations. For example, the reservation cost of time and/or usage intervals, e.g., per hour, per GB of I/O. Thus, 2 it might not be cost efficient to scale-in public cloud applications an Amazon EC2 VM service depends both on its type, and at any moment in time, without considering their billing cycles. its storage configuration, a VM optimized for high I/O costing In this work we aid developers of scalable applications for more than a regular one.
    [Show full text]