DOCSLIB.ORG

Powering Artificial Intelligence with High-Performance Computing

Total Page：16

File Type：pdf, Size：1020Kb

Download full-text PDF Read full-text
Powering Artificial Intelligence with High-Performance Computing White Paper Mission Critical Computing Powering Artificial Intelligence with High-Performance Computing September 2020 SUSE.com Powering Artificial Intelligence with 2020 High-Performance Computing Your business might be starting to recognize that an Today’s HPC systems are based on Linux clusters HPC infrastructure is vital to supporting the AI and running on industry-standard x86 hardware—the analytics applications you need to keep up with the same kind of Linux clusters enterprises are likely al- huge volumes of data and processing to compete in ready using for their big data and cloud architectures. today’s world. Updating your current IT infrastruc- ture to enable HPC to run AI workloads is vital to The HPC community is making it easier than ever to staying competitive in the current fast paced and implement a software-defined HPC infrastructure on changing environment. This paper will discuss the the hardware you already have. The OpenHPC group seven steps necessary to enable an HPC environ- and its members work to simplify the task of adopt- ment in your data center. ing HPC techniques and technologies and applying them to enterprise tasks. For this discussion, we’re broadly defining AI as any system that is trying to solve the kind of problem— How to Bring HPC to Your Organization such as recognizing patterns—that human thinking Here are seven things to keep in mind when con- does. AI encompasses machine learning (ML) and sidering how your organization is going to do AI with Deep Learning (DL) techniques. ML is a process in HPC. which curated examples are used to train a com- puter to recognize patterns. Deep learning is a PLAN YOUR WORKLOAD branch of ML that uses digital neural networks to Your obvious first step is to decide what you are going create systems that learn on their own. to run on your HPC cluster. Consider not only what you will start with, but also how you expect to scale AI applications are leveraged to accelerate competi- the application or applications. tive advantage in many industries: MPI vs Throughput Workloads - What type of work- • Management by exception load will you run? HPC workloads can be broadly • Marketing divided into two categories: parallel distributed (MPI) • Banking workloads and throughput workloads. • Finance • Agriculture Message Passing Interface (MPI) applications are par- • Healthcare allel distributed applications that consist of multiple • Gaming concurrently running processes that communicate • Space Exploration with each other with intensity. • Autonomous Vehicles • Personal assistants (Chat bots) Throughput workloads often require many tasks to • Social Media be run to complete a specific job, with each task running on its own without communication between Optimizing Your Environment to Enable HPC for tasks. One example of a throughput workload would AI Workloads be the rendering the frames of digital content. Each AI systems require scalable computing power. As frame can be computed on its own as well as in the volumes of data grow, so do training times and parallel. computational requirements. There is a lot of information out there on what type HPC environments today no longer imply bespoke of workload fits each environment and application. supercomputers built for government and academic research institutions. In fact, enterprises can build Virtualized vs Bare Metal – Historically HPC work- high-performance infrastructures while dealing with loads have been run on bare metal systems, but this far fewer vendors and far less customization than is starting to change as virtualized environments are most realize. becoming more advanced and offer more value to the enterprise and HPC environments. 2 SUSE.com Powering Artificial Intelligence with 2020 High-Performance Computing There are a number of benefits to running in a vir- and predictably with a wide range of configuration tual environment these days around heterogeneity, and connectivity options. For AI applications, Dell of- security, and flexibility. Many workloads have been fers several purpose-built platforms that feature high successfully virtualized to match the performance numbers of GPUs and dense compute environments: of bare-metal environments. • The PowerEdge C4140 Server is a two-proces- These are just some of the points that need con- sor server with up to four NVIDIA V100 GPUs in sideration when starting to think about your HPC just 1U! This server has a patented interleaved environment. Some others include resiliency, re- GPU design to optimize both space and air- dundancy, and performance. flow for maximum compute performance. The PowerEdge C4140 is available with NVIDIA UPDATING YOUR HARDWARE NVLINK™ direct GPU-to-GPU interconnect de- Your next step is to determine your hardware needs. signed to speed communication between GPUs As mentioned above, HPC can be implemented to- an order of magnitude faster than PCIe. day as a software-defined infrastructure on top of • The PowerEdge C6420 Server has up to four a standard x86 hardware cluster. independent twoprocessor servers in 2U with up to 16 DIMMs, up to 24x 2.5" hard drives and Some of the base components you will need when M.2 boot storage. It’s also available with direct assembling your HPC cluster are the following contact liquid cooling (DCLC) to support higher- wattage processors for increased performance, • Management nodes energy efficiency and racklevel density. • Compute nodes • The DSS 8440 machine learning server is a two- • Network Interconnects processor 4U server with up to 16 GPU accel- • Storage eratros and up to 10 local NVMe and SATA drives for optimized access to data. The DSS 8440 has Networking and Storage will be covered a little later, an open architecture, based on industry stan- but now we want to focus on the Management and dard PCIe fabric, allowing for customization of compute nodes and what options you have. Your accelerators, storage options and network cards. basic HPC cluster consists of at least one man- agement or login node connected to a network of SEGMENTING YOUR NETWORK multiple compute nodes. There may be multiple If you already have a Linux cluster, your existing management nodes used to run cluster-wide ser- top-of rack switches, and Ethernet network may vices, such as monitoring, workloads and storage suffice for your HPC infrastructure. More advanced usage. This depends on the size of the cluster. The HPC interconnects can be utilized to solve specific login nodes are a key component to allow users throughput issues. The diagram below shows how to access the system. Users submit their jobs us- you might segment your network traffic based on ing these login nodes to the worker nodes via a the network capacity of each type of interconnect. workload scheduler. As technology has improved over the years, the components to building a clus- User Login Data transfer Head Compute ter have changed. These days multicore Intel x86- system nodes nodes nodes nodes architecture systems with varying amounts of cores and memory are commonly used for worker nodes. 1 Gb Ethernet Hand ware mangement Dell provides several great options for both man- 1 Gb Ethernet LAN Access 1 Gb Ethernet Internodal traffic agement and compute nodes. Dell Technologies Non-compute storage 10 Gb Ethernet Application storage Database storage PowerEdge servers deliver high performance com- Infiniband Compute storage puting with the latest processors, accelerators, 40 Gb Ethernet Science DMZ storage memory and NVMe storage. You can scale efficiently 3 SUSE.com Powering Artificial Intelligence with 2020 High-Performance Computing Dell EMC switches an exascale future. It packs up to 90 hot service- Dell EMC PowerSwitch S5200 ON Series Switches able 3.5 inch drives in 4U. Available with either one provide state of the art, high density open net- or two server nodes, the DSS 7000 can deliver up working 25GbE top-of-rack and 100GbE spine/leaf to 1.26PB of storage to tackle demanding storage switches to meet the growing demands of today’s environments. HPC/AI compute and storage traffic. These combined with SUSE Enterprise Storage Dell EMC Networking Z9100 ON Series Switches are product will provide the most flexible and robust 10/25/40/50/100GbE fixed switches for high perfor- storage solution for your HPC environment running mance computing environments. With 32 ports of AI workloads. SUSE Enterprise Storage is a soft- 100GbE, 64 ports of 50GbE, 32 ports of 40GbE, 128 ware-defined storage solution powered by Ceph ports of 25GbE or 128 ports 10GbE and two SFP+ designed to help enterprises manage ever-growing ports of 10GbE/1GbE/100MbE, you can conserve data sets. rack space and simplifying migration to 100Gbps. SELECT THE RIGHT MIDDLEWARE Mellanox InfiniBand Much of the magic in HPC happens at the middle- Mellanox® SB7800 series Switch IB 2™ InfiniBand® ware layer, where you enable parallel computing EDR 100Gb/s Switches deliver high bandwidth with through tools such as workload schedulers and low latency for the highest server efficiency and ap- Message Passing Interfaces (MPIs). Luckily, many of plication productivity — ideal for HPC applications. the tools that once came from various vendors are You can get 36 ports at 100Gb/s per port, and can now available as open source. scale out to hundreds of nodes. All the tools you need are packaged together with Gen Z Consortium SUSE® Linux Enterprise Server for High Performance Dell is a founding member of the Gen Z Consortium, Computing - a highly scalable, high performance dedicated to creating a next generation intercon- open-source operating system designed to utilize nect that will bridge existing solutions while en- the power of parallel computing. abling unbounded innovation. SUSE Linux Enterprise for High-Performance Utilize Innovative Storage Solutions Computing comes with several preconfigured sys- Unprecedented growth in the amount of data col- tem roles for HPC. These roles provide a set of lected for analytics, artificial intelligence and other preselected packages as well as an installation high-performance computing makes fast, scalable workflow that will configure the systems to make and resilient storage an imperative.
Load more

Recommended publications
  • Slurm Overview and Elasticsearch Plugin
    Slurm Workload Manager Overview SC15 Alejandro Sanchez [email protected] Copyright 2015 SchedMD LLC http://www.schedmd.com Slurm Workload Manager Overview ● Originally intended as simple resource manager, but has evolved into sophisticated batch scheduler ● Able to satisfy scheduling requirements for major computer centers with use of optional plugins ● No single point of failure, backup daemons, fault-tolerant job options ● Highly scalable (3.1M core Tianhe-2 at NUDT) ● Highly portable (autoconf, extensive plugins for various environments) ● Open source (GPL v2) ● Operating on many of the world's largest computers ● About 500,000 lines of code today (plus test suite and documentation) Copyright 2015 SchedMD LLC http://www.schedmd.com Enterprise Architecture Copyright 2015 SchedMD LLC http://www.schedmd.com Architecture ● Kernel with core functions plus about 100 plugins to support various architectures and features ● Easily configured using building-block approach ● Easy to enhance for new architectures or features, typically just by adding new plugins Copyright 2015 SchedMD LLC http://www.schedmd.com Elasticsearch Plugin Copyright 2015 SchedMD LLC http://www.schedmd.com Scheduling Capabilities ● Fair-share scheduling with hierarchical bank accounts ● Preemptive and gang scheduling (time-slicing parallel jobs) ● Integrated with database for accounting and configuration ● Resource allocations optimized for topology ● Advanced resource reservations ● Manages resources across an enterprise Copyright 2015 SchedMD LLC http://www.schedmd.com
    [Show full text]
  • Intel® SSF Reference Design: Intel® Xeon Phi™ Processor, Intel® OPA
    Intel® Scalable System Framework Reference Design Intel® Scalable System Framework (Intel® SSF) Reference Design Cluster installation for systems based on Intel® Xeon® processor E5-2600 v4 family including Intel® Ethernet. Based on OpenHPC v1.1 Version 1.0 Intel® Scalable System Framework Reference Design Summary This Reference Design is part of the Intel® Scalable System Framework series of reference collateral. The Reference Design is a verified implementation example of a given reference architecture, complete with hardware and software Bill of Materials information and cluster configuration instructions. It can confidently be used “as is”, or be the foundation for enhancements and/or modifications. Additional Reference Designs are expected in the future to provide example solutions for existing reference architecture definitions and for utilizing additional Intel® SSF elements. Similarly, more Reference Designs are expected as new reference architecture definitions are introduced. This Reference Design is developed in support of the specification listed below using certain Intel® SSF elements: • Intel® Scalable System Framework Architecture Specification • Servers with Intel® Xeon® processor E5-2600 v4 family processors • Intel® Ethernet • Software stack based on OpenHPC v1.1 2 Version 1.0 Legal Notices No license (express or implied, by estoppel or otherwise) to any intellectual property rights is granted by this document. Intel disclaims all express and implied warranties, including without limitation, the implied warranties of merchantability, fitness for a particular purpose, and non-infringement, as well as any warranty arising from course of performance, course of dealing, or usage in trade. This document contains information on products, services and/or processes in development. All information provided here is subject to change without notice.
    [Show full text]
  • SUSE Linux Enterprise High Performance Computing 15 SP3 Administration Guide Administration Guide SUSE Linux Enterprise High Performance Computing 15 SP3
    SUSE Linux Enterprise High Performance Computing 15 SP3 Administration Guide Administration Guide SUSE Linux Enterprise High Performance Computing 15 SP3 SUSE Linux Enterprise High Performance Computing Publication Date: September 24, 2021 SUSE LLC 1800 South Novell Place Provo, UT 84606 USA https://documentation.suse.com Copyright © 2020–2021 SUSE LLC and contributors. All rights reserved. Permission is granted to copy, distribute and/or modify this document under the terms of the GNU Free Documentation License, Version 1.2 or (at your option) version 1.3; with the Invariant Section being this copyright notice and license. A copy of the license version 1.2 is included in the section entitled “GNU Free Documentation License”. For SUSE trademarks, see http://www.suse.com/company/legal/ . All third-party trademarks are the property of their respective owners. Trademark symbols (®, ™ etc.) denote trademarks of SUSE and its aliates. Asterisks (*) denote third-party trademarks. All information found in this book has been compiled with utmost attention to detail. However, this does not guarantee complete accuracy. Neither SUSE LLC, its aliates, the authors nor the translators shall be held liable for possible errors or the consequences thereof. Contents Preface vii 1 Available documentation vii 2 Giving feedback vii 3 Documentation conventions viii 4 Support x Support statement for SUSE Linux Enterprise High Performance Computing x • Technology previews xi 1 Introduction 1 1.1 Components provided 1 1.2 Hardware platform support 2 1.3 Support
    [Show full text]
  • Version Control Graphical Interface for Open Ondemand
    VERSION CONTROL GRAPHICAL INTERFACE FOR OPEN ONDEMAND by HUAN CHEN Submitted in partial fulfillment of the requirements for the degree of Master of Science Department of Electrical Engineering and Computer Science CASE WESTERN RESERVE UNIVERSITY AUGUST 2018 CASE WESTERN RESERVE UNIVERSITY SCHOOL OF GRADUATE STUDIES We hereby approve the thesis/dissertation of Huan Chen candidate for the degree of Master of Science Committee Chair Chris Fietkiewicz Committee Member Christian Zorman Committee Member Roger Bielefeld Date of Defense June 27, 2018 TABLE OF CONTENTS Abstract CHAPTER 1: INTRODUCTION ............................................................................ 1 CHAPTER 2: METHODS ...................................................................................... 4 2.1 Installation for Environments and Open OnDemand .............................................. 4 2.1.1 Install SLURM ................................................................................................. 4 2.1.1.1 Create User .................................................................................... 4 2.1.1.2 Install and Configure Munge ........................................................... 5 2.1.1.3 Install and Configure SLURM ......................................................... 6 2.1.1.4 Enable Accounting ......................................................................... 7 2.1.2 Install Open OnDemand .................................................................................. 9 2.2 Git Version Control for Open OnDemand
    [Show full text]
  • Extending SLURM for Dynamic Resource-Aware Adaptive Batch
    1 Extending SLURM for Dynamic Resource-Aware Adaptive Batch Scheduling Mohak Chadha∗, Jophin John†, Michael Gerndt‡ Chair of Computer Architecture and Parallel Systems, Technische Universitat¨ Munchen¨ Garching (near Munich), Germany Email: [email protected], [email protected], [email protected] their resource requirements at runtime due to varying computa- Abstract—With the growing constraints on power budget tional phases based upon refinement or coarsening of meshes. and increasing hardware failure rates, the operation of future A solution to overcome these challenges and efficiently utilize exascale systems faces several challenges. Towards this, resource awareness and adaptivity by enabling malleable jobs has been the system components is resource awareness and adaptivity. actively researched in the HPC community. Malleable jobs can A method to achieve adaptivity in HPC systems is by enabling change their computing resources at runtime and can signifi- malleable jobs. cantly improve HPC system performance. However, due to the A job is said to be malleable if it can adapt to resource rigid nature of popular parallel programming paradigms such changes triggered by the batch system at runtime [6]. These as MPI and lack of support for dynamic resource management in batch systems, malleable jobs have been largely unrealized. In resource changes can either increase (expand operation) or this paper, we extend the SLURM batch system to support the reduce (shrink operation) the number of processors. Malleable execution and batch scheduling of malleable jobs. The malleable jobs have shown to significantly improve the performance of a applications are written using a new adaptive parallel paradigm batch system in terms of both system and user-centric metrics called Invasive MPI which extends the MPI standard to support such as system utilization and average response time [7], resource-adaptivity at runtime.
    [Show full text]
  • Distributed HPC Applications with Unprivileged Containers Felix Abecassis, Jonathan Calmels GPUNVIDIA Computing Beyond Video Games
    Distributed HPC Applications with Unprivileged Containers Felix Abecassis, Jonathan Calmels GPUNVIDIA Computing Beyond video games VIRTUAL REALITY SCIENTIFIC COMPUTING MACHINE LEARNING AUTONOMOUS MACHINES GPU COMPUTING 2 Infrastructure at NVIDIA DGX SuperPOD GPU cluster (Top500 #20) x96 3 NVIDIA Containers Supports all major container runtimes We built libnvidia-container to make it easy to run CUDA applications inside containers We release optimized container images for each of the major Deep Learning frameworks every month We use containers for everything on our HPC clusters - R&D, official benchmarks, etc Containers give us portable software stacks without sacrificing performance 4 Typical cloud deployment e.g. Kubernetes Hundreds/thousands of small nodes All applications are containerized, for security reasons Many small applications running per node (e.g. microservices) Traffic to/from the outside world Not used for interactive applications or development Advanced features: rolling updates with rollback, load balancing, service discovery 5 GPU Computing at NVIDIA HPC-like 10-100 very large nodes “Trusted” users Not all applications are containerized Few applications per node (often just a single one) Large multi-node jobs with checkpointing (e.g. Deep Learning training) Little traffic to the outside world, or air-gapped Internal traffic is mostly RDMA 6 Slurm Workload Manager https://slurm.schedmd.com/slurm.html Advanced scheduling algorithms (fair-share, backfill, preemption, hierarchical quotas) Gang scheduling: scheduling and
    [Show full text]
  • Cluster Deployment
    INAF-OATs Technical Report 222 - INCAS: INtensive Clustered ARM SoC - Cluster Deployment S. Bertoccoa, D. Goza, L. Tornatorea, G. Taffonia aINAF-Osservatorio Astronomico di Trieste, Via G. Tiepolo 11, 34131 Trieste - Italy Abstract The report describes the work done to deploy an Heterogeneous Computing Cluster using CPUs and GPUs integrated on a System on Chip architecture, in the environment of the ExaNeSt H2020 project. After deployment, this cluster will provide a testbed to validate and optimize several HPC codes developed to be run on heterogeneous hardware. The work done includes: a technology watching phase to choose the SoC best fitting our requirements; installation of the Operative System Ubuntu 16.04 on the choosen hardware; cluster layout and network configuration; software installation (cluster management software slurm, specific software for HPC programming). Keywords: SoC, Firefly-RK3399, Heterogeneous Cluster Technology HCT, slurm, MPI, OpenCL, ARM 1. Introduction The ExaNeSt H2020 project aims at the design and development of an exascale ready supercomputer with a low energy consumption profile but able to support the most demanding scientific and technical applica- tions. The project will produce a prototype based on hybrid hardware (CPUs+accelerators). Astrophysical codes are playing a fundamental role to validate this exascale platform. A preliminary work has been done to Email addresses: [email protected] ORCID:0000-0003-2386-623X (S. Bertocco), [email protected] ORCID:0000-0001-9808-2283 (D. Goz), [email protected] ORCID:0000-0003-1751-0130 (L. Tornatore), [email protected] ORCID:0000-0001-7011-4589 (G. Taffoni) Preprint submitted to INAF August 3, 2018 port on the heterogeneous platform a state-of-the-art N-body code (called ExaHiGPUs) [4].
    [Show full text]
  • Sansa - the Supercomputer and Node State Architecture
    SaNSA - the Supercomputer and Node State Architecture Neil Agarwal∗y, Hugh Greenberg∗, Sean Blanchard∗, and Nathan DeBardeleben∗ ∗ Ultrascale Systems Research Center Los Alamos National Laboratory High Performance Computing Los Alamos, New Mexico, 87545 Email: fhng, seanb, [email protected] y University of California, Berkeley Berkeley, California, 94720 Email: [email protected] Abstract—In this work we present SaNSA, the Supercomputer vendor, today there exists relatively poor fine-grained informa- and Node State Architecture, a software infrastructure for histor- tion about the state of individual nodes of a cluster. Operators ical analysis and anomaly detection. SaNSA consumes data from and technicians of data centers have dashboards which actively multiple sources including system logs, the resource manager, scheduler, and job logs. Furthermore, additional context such provide real time feedback about coarse-grained information as scheduled maintenance events or dedicated application run such as whether a node is up or down. times for specific science teams can be overlaid. We discuss how The Slurm Workload Manager, which runs on most of the this contextual information allows for more nuanced analysis. U.S. Department of Energy’s clusters, has one view of the SaNSA allows the user to apply arbitrary attributes, for instance, state of the nodes in the system. Slurm uses this to know positional information where nodes are located in a data center. We show how using this information we identify anomalous whether nodes are available for scheduling, jobs are running, behavior of one rack of a 1,500 node cluster. We explain the and when they complete. However, there are other views such design of SaNSA and then test it on four open compute clusters as the node’s own view as expressed via system logs it reports at LANL.
    [Show full text]
  • Intelligent Colocation of HPC Workloads?,??
    Intelligent colocation of HPC workloads?;?? Felippe Vieira Zacariasa,b,c, Vinicius Petruccia,e, Rajiv Nishtalad, Paul Carpenterc, Daniel Mosse´e aUniversidade Federal da Bahia bUniversitat Polit`ecnicade Catalunya cBarcelona Supercomputing Center dCoop, Norway/Norwegian University of Science and Technology, Norway eUniversity of Pittsburgh Abstract Many HPC applications suffer from a bottleneck in the shared caches, instruction execution units, I/O or memory bandwidth, even though the remaining resources may be underutilized. It is hard for developers and runtime systems to ensure that all critical resources are fully exploited by a single application, so an attractive technique for increasing HPC system utilization is to colocate multiple applications on the same server. When applications share critical resources, however, contention on shared resources may lead to reduced application performance. In this paper, we show that server efficiency can be improved by first modeling the expected performance degradation of colocated applications based on measured hardware performance counters, and then exploiting the model to determine an optimized mix of colocated applications. This paper presents a new intelligent resource manager and makes the following contributions: (1) a new machine learning model to predict the performance degradation of colocated applications based on hardware counters and (2) an intelligent scheduling scheme deployed on an existing resource manager to enable application co-scheduling with minimum performance degradation. Our results show that our approach achieves performance improvements of 7 % (avg) and 12 % (max) compared to the standard policy commonly used by existing job managers. Keywords: Resource manager, HPC systems, Machine learning, Colocation, Performance Characterization, Performance counters 1. Introduction is key to keeping up with increased computational demands.
    [Show full text]
  • Winter 2016 Vol
    ;login WINTER 2016 VOL. 41, NO. 4 : & Your Cores Are Slacking Off Jean-Pierre Lozi, Baptiste Lepers, Justin Funston, Fabien Gaud, Vivien Quéma, and Alexandra Fedorova & BeyondCorp Part III: The Access Proxy Luca Cittadini, Batz Spear, Betsy Beyer, and Max Saltonstall & OpenLambda for Microservices Scott Hendrickson, Stephen Sturdevant, Edward Oakes, Tyler Harter, Venkateshwaran Venkataramani, Andrea Arpaci-Dusseau, and Remzi Arpaci-Dusseau & Tuning ZFS for Databases Allan Jude and Michael W. Lucas & Interview with Fyodor Vaskovich Rik Farrow Columns Understanding Python Metaclasses David Beazley Monitoring Your Monitoring Systems Dave Josephsen Forecasting the Weather with Dark Sky David N. Blank-Edelman Extending Go Applications with Exec Plugins Kelsey Hightower Rising Tide of IoT Devices Dan Geer Idiot-Blockers for the Internet Robert G. Ferrell UPCOMING EVENTS LISA16 USENIX ATC ’17: 2017 USENIX Annual Technical December 4–9, 2016, Boston, MA, USA Conference www.usenix.org/lisa16 July 12-14, 2017, Santa Clara, CA, USA Co-located with LISA16 Submissions due February 7, 2017 www.usenix.org/atc17 SESA ’16: 2016 USENIX Summit for Educators in System Administration Co-located with USENIX ATC ’17 December 6, 2016 HotCloud ’17: 9th USENIX Workshop on Hot Topics www.usenix.org/sesa16 in Cloud Computing July 10-11, 2017, Santa Clara, CA, USA USENIX Journal of Education in System Administration Submissions due March 14, 2017 (JESA) www.usenix.org/hotcloud17 Published in conjunction with SESA www.usenix.org/jesa HotStorage ’17: 9th USENIX Workshop
    [Show full text]
  • Virtual Machine Checkpoint Storage and Distribution for Simuboost
    Virtual Machine Checkpoint Storage and Distribution for SimuBoost Master Thesis by Bastian Eicher at the Department of Computer Science Operating Systems Group Karlsruhe Institute of Technology Supervisor: Prof. Dr. Frank Bellosa Supervising Research Assistant:Dipl.-Inform. Marc Rittinghaus Created during: April 1st 2015 – September 4th 2015 KIT – Universität des Landes Baden-Württemberg und nationales Forschungszentrum in der Helmholtz-Gemeinschaft www.kit.edu iii I hereby declare that this thesis is my own original work which I created without illegitimate help by others, that I have not used any other sources or resources than the ones indicated and that due acknowledgment is given where reference is made to the work of others. Karlsruhe, September 4th 2015 iv Abstract While full-system simulation enables detailed analysis of workloads it is much slower than hardware-assisted virtualization, with slowdown factors ranging from 30 to 1000. The SimuBoost concept aims to combine the benefits of full-system simulation and virtualization. Checkpoints of a virtual machine’s state are cre- ated in regular intervals and used to seed parallelized distributed simulations. To reach an optimal degree of parallelization, checkpoints have to be created in short intervals with low downtimes. In this thesis we evaluate the viability of the SimuBoost concept. We improve an existing checkpointing prototype to minimize the downtime by replacing the un- derlying storage mechanism and performing asynchronous deduplication. We also implement a cluster-distribution solution in order to measure the actual speedup achievable by SimuBoost. The evaluation shows that the new storage mechanism and performing deduplica- tion asynchronously greatly reduce the downtime caused by checkpointing.
    [Show full text]
  • The Crossroads of Cloud And
    Table of Contents Introduction 1 OpenStack and Virtualised HPC 5 OpenStack and HPC Network Fabrics 14 OpenStack and High-Performance Data 25 OpenStack and HPC Workload Management 38 OpenStack and HPC Infrastructure Management 52 OpenStack and Research Cloud Federation 67 Summary 77 AUTHORS “OpenStack and Virtualised HPC”, “OpenStack and HPC Network Fabrics”, “OpenStack and High-Performance Data”, “OpenStack and HPC Workload Management” and “OpenStack and HPC Infrastructure Management” by Stig Telfer, CTO, StackHPC Ltd. with the support of Cambridge University. “OpenStack and Research Cloud Federation” by Enol Fernández, Cloud Technologist, EGI. CONTRIBUTIONS, GUIDANCE AND FEEDBACK OPENSTACK AND VIRTUALISED HPC: Blair Bethwaite, Senior HPC Consultant, Monash University Dr. Xiaoyi Lu, Senior Research Associate, NOWLAB, The Ohio State University Dhabaleswar K. (DK) Panda, Professor and University Distinguished Scholar, NOWLAB, The Ohio State University OPENSTACK AND HPC NETWORK FABRICS: Paul Browne, OpenStack Engineer, Cambridge University Dr. Xiaoyi Lu, Senior Research Associate, NOWLAB, The Ohio State University Jonathan Mills, Senior HPC Engineer, CSC @ NASA Goddard Space Flight Center Dhabaleswar K. (DK) Panda, Professor and University Distinguished Scholar, NOWLAB, The Ohio State University OPENSTACK AND HIGH-PERFORMANCE DATA: Dr. Tom Connor, Senior Lecturer, Cardiff University and the CLIMB collaboration Vincent Ferretti, Senior Principal Investigator & Associate Director, Bioinformatics Software Development, Ontario Institute for Cancer
    [Show full text]