DOCSLIB.ORG

Exploring a Multithreaded Methodology to Implement a Network Communication Protocol on the Cyclops-64 Multithreaded Architecture

Total Page：16

File Type：pdf, Size：1020Kb

Download full-text PDF Read full-text
Exploring a Multithreaded Methodology to Implement a Network Communication Protocol on the Cyclops-64 Multithreaded Architecture Exploring a Multithreaded Methodology to Implement a Network Communication Protocol on the Cyclops-64 Multithreaded Architecture Ge Gan, Ziang Hu, Juan del Cuvillo, Guang R. Gao University of Delaware Dept. of Electrical and Computer Engineering Newark, DE. 19716 USA {gan,hu,jcuvillo,ggao}@capsl.udel.edu C64 Node Abstract C64 Chip icache−5P icache−5P icache−5P icache−5P TU SP GM icache−5P FPU TU SP GM icache−5P icache−5P The IBM Cyclops-64 (C64) chip employs a multi- icache−5P icache−5P ...... icache−5P icache−5P TU SP GM icache−5P 6 threaded architecture that integrates a large number of FPU TU SP GM icache−5P icache−5P icache−5P hardware thread units on a single chip. A cellular super- TU SP GM IC−Glue IC−Glueicache−5P IC−Glue IC−Glue FPU TU SP GM DDR2 Controller computer is being developed based on a 3D-mesh connec- TU SP GM DDR2 Controller FPU TU SP GM 96−Port Crossbar Switch DDR2 Controller 1G Offchip DRAM tion of the C64 chips. This paper introduces the Cyclops TU SP GM Datagram Protocol (CDP) developed for the C64 super- FPU TU SP GM DDR2 Controller TU SP GM 2 3D−mesh A−Switch FPU TU SP GM computer system. CDP is inspired by the TCP/IP protocol, Gigabit Ethernet TU SP GM 1 345 80 Host Interface IDE FPU TU SP GM yet simpler and more compact. The implementation of CDP FPGA Control Network leverages the abundant hardware thread-level parallelism TU: Thread Unit SP: Scratchpad Memory FPU: Floating Point Unit GM: Global Memory provided by the C64 multithreaded architecture. The main contributions of this paper are: (1) We have Figure 1. Cyclops-64 Node completed a design and implementation of CDP that is used as the fundamental communication infrastructure for the C64 supercomputer system. (2) CDP successfully ex- ics and the off-chip memory, becomes the building block ploits the massive thread-level parallelism provided on the (i.e. the C64 node) of the C64 supercomputer system. See C64 hardware, achieving good performance scalability; Figure 2. The C64 supercomputer system consists of tens of (3) CDP is quite efficient. Its peak throughput reaches thousands of C64 nodes that are connected by the 3D-mesh 884Mbps on the Gigabit Ethernet, even it is running at network and the Gigabit Ethernet and can provide comput- the user-level on a single-processor Linux machine; (4) Ex- ing power at the Petaflops level. tensive application test cases are passed and no reliability To interconnect the two different subnetworks in the C64 problems have been reported. supercomputer, we have designed the Cyclops Datagram Protocol (CDP). CDP is a projection of the conventional network communication protocol (TCP/IP) to the modern 1. Introduction C64 multithreaded architecture. It is a datagram-based, connection-oriented communication protocol that supports Cyclops-64 (C64) is a multithreaded architecture devel- reliable and full-duplex data transfer. We have implemented oped at the IBM T.J. Watson research center [4]. It is the lat- the very popular BSD socket API in CDP. This provides a est version of the Cyclops cellular architecture that employs user-friendly programming environment for the C64 system a unique multiprocessor-on-a-chip design [1] that integrates or application programmers. a large number of thread execution units, main memory We have implemented the CDP protocol on the C64 banks, and communication hardware on a single chip. See thread virtual machine (TVM) [3]. The C64 thread vir- Figure 1. The C64 chip, together with the host control log- tual machine is a lightweight runtime system developed for the C64 chip. It provides the mechanism to map software 1-4244-0910-1/07/$20.00 c 2007 IEEE. threads directly onto the C64 hardware thread units. It also The A-switch interface of the chip connects the C64 node to its six neighbors in the 3D-mesh network. In every CPU cycle, A-switch can transfer one double word (8 bytes) .... in one direction. The 3D-mesh may scale up to several ten thousands of nodes, which becomes a powerful parallel 3D−mesh Gigabits Ethernet I/O node computing engine. The 3D-mesh computing engine is at- compute node tached to the host system via Gigabit Ethernet and becomes host node the C64 supercomputer. See Figure 2. The whole C64 sys- Figure 2. Cyclops-64 Supercomputer tem is designed to provide computing power at Petaflops level. It is targeted at applications that are highly paralleliz- able and require enormous amount of computation. provides a familiar and efficient programming interface for Given the C64 multithreaded architecture and the C64 the C64 system programmers. Currently the C64 hardware supercomputer organization, we are interested in two ques- is still under development, so the C64 thread virtual ma- tions regarding the design and implementation of CDP: chine is running on the C64 FAST simulator [2]. • Is it possible to implement CDP in a way such that it We have explored a multithreaded methodology in the can utilize the massive thread-level parallelism on the development of the CDP protocol. A fine-grain thread li- C64 hardware and achieve good performance scalabil- brary called TiNy Thread (TNT) library [3] is used to im- ity? plement the CDP protocol. The TNT thread library is part of the C64 Thread Virtual Machine and implements the C64 • Is the communication protocol we developed for the fine-grain thread model [3]. C64 architecture an efficient one? We have evaluated the performance of CDP through micro-benchmarking. From the experimental results, we CDP Node CDP Node have two observations: (1) The multithreaded methodology used in the implementation of CDP is very successful. It ef- .. .. .. .. fectively exploits the massive thread-level parallelism pro- user thread user thread user thread user thread vided on the C64 hardware and achieves good performance cdp receiving thread cdp receiving thread cdp timer thread cdp receiving thread cdp receiving thread cdp timer thread cdplib cdplib cdplib cdplib scalability. The speedup of a CDP test program can reach 82.55 after using 128 receiving threads. (2) As a commu- 3D−mesh network / Gigabit Ethernet nication protocol, CDP is efficient. The peak throughput of the user-level CDP (implemented by Pthread) is 884Mbps Figure 3. CDP Multithreaded Implementation on the Gigabit Etherent. In the next section, we will first introduce the necessary In order to answer these questions, we came up a multi- background of the C64 architecture. Then, we will formu- threaded solution, shown in Figure 3. Briefly, the CDP late our problem and give a brief introduction to the solu- program consists of a set of TNT threads [3]: the receiv- tion. ing threads, the timer thread, and the user threads. These threads cooperate with each other to implement the full 2 Problem Formulation and Solution functions of the CDP protocol. A fine-grain lock algorithm is proposed to improve the parallelism among these threads. Section 4 will give a detailed description on the CDP imple- A C64 chip has 80 “processors”, which are connected mentation. to a 96-port crossbar network. See Figure 1. Each proces- The rest of the paper is organized as follows. Section 3 sor consists of two thread units, one floating point unit, and briefly introduces the CDP communication protocol. Sec- two SRAM memory banks (32KB each). A thread unit is a tion 4 discusses the multithreaded implementation of CDP. 64-bit, single issue, in-order RISC core operating at clock Section 5 presents the experimental results and analysis. rate of 500MHz. The execution on the thread unit is not Section 6 introduces some related works. Section 7 is our preemptable. A 32KB instruction cache is shared among conclusion. We will talk a little about our future work in five processors. The chip has no cache for data. Instead, section 8. a portion of the SRAM memory bank can be configured as scratchpad memory (SP), which is a fast temporary stor- age that can be used to exploit locality under software con- 3 CDP Protocol trol. All of the remaining part of the SRAM form the global memory (GM) and is uniformly addressable from all thread CDP is inspired by TCP/IP, yet simpler and more com- units. The C64 chip does not support virtual memory. pact. See Figure 5. Such a design is based on the consider- 2 IP Header ation that both the C64 architecture and the network topol- ver IHL ToS Total Length identification fragment offset ogy of the C64 supercomputer are simple. We will briefly TTL protocol Header Checksum introduce the CDP protocol in this section and discuss the Source Address CDP Header Destination Address destination node destination port protocol implementation problems in the next section. Options (optional) source node source port sequence number TCP Header acknowledgment number source port destination port flags total length 3.1 Overview sequence number acknowledgment number flags window Checksum Urgent Pointer Figure 4 shows the position of CDP in the protocol stack. Options (optional) According to the OSI reference model, CDP corresponds to the Transport layer plus the Network layer. This implies Figure 5. CDP Packet Header Format that CDP should implement the main functions (or at least some) of these two layers that are specified in the OSI refer- ence model. Here are the main features of CDP: (1) CDP is As for the CDP connection, the finite state automata used a datagram-based, connection-oriented communication pro- to direct the connection state transition is shown in Figure tocol; (2) it is reliable and supports timeout retransmission; 6. This finite state automata is similar to the one used in (3) it uses sliding-window based flow control mechanism starting point to avoid network traffic congestion; (4) it provides a full- CLOSED server: cdp_listen() duplex service to the application layer; (5) it has imple- client: cdp_connect() LISTEN send: SYN mented the very familiar BSD Socket programming inter- recv: SYN; passive open send: SYN,ACK faces for the CDP program developers.
Load more

Recommended publications
  • System Trends and Their Impact on Future Microprocessor Design
    IBM Research System Trends and their Impact on Future Microprocessor Design Tilak Agerwala Vice President, Systems IBM Research System Trends and their Impact | MICRO 35 | Tilak Agerwala © 2002 IBM Corporation IBM Research Agenda System and application trends Impact on architecture and microarchitecture The Memory Wall Cellular architectures and IBM's Blue Gene Summary System Trends and their Impact | MICRO 35 | Tilak Agerwala © 2002 IBM Corporation IBM Research TRENDSTRENDS Microprocessors < 10 GHz in systems 64-256 Way SMP 65-45nm, Copper, Highest performance SOI Best MP Scalability 1-2 GHz Leading edge process technology 4-8 Way SMP RAS, virtualization ~100nm technology 10+ of GHz 4-8 Way SMP SMP/Large 65-45nm, Copper, SOI Systems Highest Frequency Cost and power Low GHz sensitive Leading edge process Uniprocessor technology Desktop ~100nm technology 2-4 GHz, Uniproc, Component-based and Game ~100nm, Copper, SOI Consoles Lowest Power / Lowest Multi MHz cost designs SoC capable Embedded Uniprocessor ASIC / Foundry ~100-200nm technologies Systems technology System Trends and their Impact | MICRO 35 | Tilak Agerwala © 2002 IBM Corporation IBM Research TRENDSTRENDS Large system application trends Traditional commercial applications Databases, transaction processing, business apps like payroll etc. The internet has driven the growth of new commercial applications New life sciences applications are commercial and high-growth Drug discovery and genetic engineering research needs huge amounts of compute power (e.g. protein folding simulations)
    [Show full text]
  • An Overview of the Blue Gene/L System Software Organization
    An Overview of the Blue Gene/L System Software Organization George Almasi´ , Ralph Bellofatto , Jose´ Brunheroto , Calin˘ Cas¸caval , Jose´ G. ¡ Castanos˜ , Luis Ceze , Paul Crumley , C. Christopher Erway , Joseph Gagliano , Derek Lieber , Xavier Martorell , Jose´ E. Moreira , Alda Sanomiya , and Karin ¡ Strauss ¢ IBM Thomas J. Watson Research Center Yorktown Heights, NY 10598-0218 £ gheorghe,ralphbel,brunhe,cascaval,castanos,pgc,erway, jgaglia,lieber,xavim,jmoreira,sanomiya ¤ @us.ibm.com ¥ Department of Computer Science University of Illinois at Urbana-Champaign Urabana, IL 61801 £ luisceze,kstrauss ¤ @uiuc.edu Abstract. The Blue Gene/L supercomputer will use system-on-a-chip integra- tion and a highly scalable cellular architecture. With 65,536 compute nodes, Blue Gene/L represents a new level of complexity for parallel system software, with specific challenges in the areas of scalability, maintenance and usability. In this paper we present our vision of a software architecture that faces up to these challenges, and the simulation framework that we have used for our experiments. 1 Introduction In November 2001 IBM announced a partnership with Lawrence Livermore National Laboratory to build the Blue Gene/L (BG/L) supercomputer, a 65,536-node machine de- signed around embedded PowerPC processors. Through the use of system-on-a-chip in- tegration [10], coupled with a highly scalable cellular architecture, Blue Gene/L will de- liver 180 or 360 Teraflops of peak computing power, depending on the utilization mode. Blue Gene/L represents a new level of scalability for parallel systems. Whereas existing large scale systems range in size from hundreds (ASCI White [2], Earth Simulator [4]) to a few thousands (Cplant [3], ASCI Red [1]) of compute nodes, Blue Gene/L makes a jump of almost two orders of magnitude.
    [Show full text]
  • Cellular Wave Computers and CNN Technology – a Soc Architecture with Xk Processors and Sensor Arrays*
    Cellular Wave Computers and CNN Technology – a SoC architecture with xK processors and sensor arrays* Tamás ROSKA1 Fellow IEEE 1. Introduction and the main theses 1.1 Scenario: Architectural lessons from the trends in manufacturing billion component devices when crossing the threshold of 100 nm feature size Preliminary proposition: The nature of fabrication technology, the nature and type of data to be processed, and the nature and type of events to be detected or „computed” will determine the architecture, the elementary instructions, and the type of algorithms needed, hence also the complexity of the solution. In view of this proposition, let us list a few key features of the electronic technology of Today and its consequences.. (i) Convergence of CMOS, NANO and OPTICAL technologies towards a cellular architecture with short and sparse wires CMOS chips: * Processors: K or M transistors on an M or G transistor die =>K processors /chip * Wires: at 180 nm or below, gate delay is smaller than wire delay NANO processors and sensors: * Mainly 2D organization of cells integrating processing and sensing * Interactions mainly with the neighbours OPTICAL devices: * parallel processing * optical correlators * VCSELs and programable SLMs, Hence the architecture should be characterized by * 2 D layers (or a layered 3D) * Cellular architecture with * mainly local and /or regular sparse wireing leading via Î a Cellular Nonlinear Network (CNN) Dynamics 1 The Faculty of Information Technology and the Jedlik Laboratories of the Pázmány University, Budapest and the Computer and Automation Institute of the Hungarian Academy of Sciences, Budapest, Hungary ([email protected], www.itk.ppke.hu) * Research supported by the Office of Naval Research, Human Frontiers of Science Program, EU Future and Emergent Technologies Program, the Hungarian Academy of Sciences, and the Jedlik Laboratories of the Pázmány University, Budapest 0-7803-9254-X/05/$20.00 ©2005 IEEE.
    [Show full text]
  • Performance Modelling and Optimization of Memory Access on Cellular Computer Architecture Cyclops64
    Performance modelling and optimization of memory access on cellular computer architecture Cyclops64 Yanwei Niu, Ziang Hu, Kenneth Barner and Guang R. Gao Department of ECE, University of Delaware, Newark, DE, 19711, USA {niu, hu, barner, ggao}@ee.udel.edu Abstract. This paper focuses on the Cyclops64 computer architecture and presents an analytical model and performance simulation results for the preloading and loop unrolling approaches to optimize the performance of SVD (Singular Value Decomposition) benchmark. A performance model for dissecting the total execu- tion cycles is presented. The data preloading using “memcpy” or hand optimized “inline” assembly code, and the loop unrolling approach are implemented and compared with each other in terms of the total number of memory access cycles. The key idea is to preload data from offchip to onchip memory and store the data back after the computation. These approaches can reduce the total memory access cycles and can thus improve the benchmark performance significantly. 1 Introduction The design concept of computer architecture over the last two decades has been mainly on the exploitation of the instruction level parallelism, such as pipelining,VLIW or superscalar architecture. For the next generation of computer architecture, hardware threading multiprocessor is becoming more and more popular. One approach of hard- ware multithreading is called CMP (Chip MultiProcessor) approach, which proposes a single chip design that uses a collection of independent processors with less resource sharing. An example of CMP architecture design is Cyclops64 [1–5], a new architec- ture for high performance parallel computers being developed at the IBM T. J. Watson Research Center and University of Delaware.
    [Show full text]
  • Focal-Plane Analog VLSI Cellular Implementation of the Boundary Contour System
    IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS—I: FUNDAMENTAL THEORY AND APPLICATIONS, VOL. 46, NO. 2, FEBRUARY 1999 327 [6] K. F. Hui and B. E. Shi, “Robustness of CNN implementations for Gabor-type image filtering,” in Proc. Asia Pacific Conf. Circuits Systems, Seoul, South Korea, Nov. 1996, pp. 105–108. [7] C. C. Enz, F. Krummenacher, and E. A. Vittoz, “An analytical MOS transistor model valid in all regions of operation and dedicated to low-voltage and low-current applications,” Anal. Integr. Circuits Signal Processing, vol. 8, no. 1, pp. 83–114, July 1995. [8] B. E. Shi and K. F. Hui, “An analog VLSI neural network for phase based machine vision,” in Advances in Neural Information Process- ing Systems 10, M. I. Jordan, M. J. Kearns, and S. A. Solla, Eds. Cambridge, MA: MIT, 1998, pp. 726–732. [9] C. A. Mead and T. Delbruck, “Scanners for visualizing activity of analog VLSI circuitry,” Anal. Integr. Circuits Signal Processing, vol. 1, no. 2, pp. 93–106, Oct. 1991. (a) Focal-Plane Analog VLSI Cellular Implementation of the Boundary Contour System Gert Cauwenberghs and James Waskiewicz Abstract—We present an analog very large scale integration (VLSI) cellular architecture implementing a version of the boundary contour system (BCS) for real-time focal-plane image processing. Inspired by neuromorphic models across the retina and several layers of visual cortex, the design integrates in each pixel the functions of phototransduction and simple cells, complex cells, hypercomplex cells, and bipole cells in each of three directions interconnected on a hexagonal grid. Analog current- mode complementary metal–oxide–semiconductor (CMOS) circuits are used throughout to perform edge detection, local inhibition, directionally (b) selective long-range diffusive kernels, and renormalizing global gain control.
    [Show full text]
  • Software-Defined Hyper-Cellular Architecture for Green and Elastic
    1 Software-Defined Hyper-Cellular Architecture for Green and Elastic Wireless Access Sheng Zhou, Tao Zhao, Zhisheng Niu, and Shidong Zhou Abstract To meet the surging demand of increasing mobile Internet traffic from diverse applications while maintaining moderate energy cost, the radio access network (RAN) of cellular systems needs to take a green path into the future, and the key lies in providing elastic service to dynamic traffic demands. To achieve this, it is time to rethink RAN architectures and expect breakthroughs. In this article, we review the state-of-art literature which aims to renovate RANs from the perspectives of control-traffic decoupled air interface, cloud-based RANs, and software-defined RANs. We then propose a software- defined hyper-cellular architecture (SDHCA) that identifies a feasible way of integrating the above three trends to enable green and elastic wireless access. We further present key enabling technologies to realize SDHCA, including separation of the air interface, green base station operations, and base station functions virtualization, followed by our hardware testbed for SDHCA. Besides, we summarize several future research issues worth investigating. I. INTRODUCTION Since their birth, cellular systems have evolved from the first generation analog systems with very low data rate to today’s fourth generation (4G) systems with more than 100 Mbps capacity to end users. However, the radio access network (RAN) architecture has not experienced arXiv:1512.04935v1 [cs.NI] 15 Dec 2015 many changes: base stations (BSs) are generally deployed and operated in a distributed fashion, Sheng Zhou, Zhisheng Niu and Shidong Zhou are with Tsinghua National Laboratory for Information Science and Technology, Dept.
    [Show full text]
  • An Overview on Cyclops-64 Architecture - a Status Report on the Programming Model and Software Infrastructure
    An Overview on Cyclops-64 Architecture - A Status Report on the Programming Model and Software Infrastructure Guang R. Gao Endowed Distinguished Professor Electrical & Computer Engineering University of Delaware [email protected] 2007/6/14 SOS11-06-2007.ppt 1 Outline • Introduction • Multi-Core Chip Technology • IBM Cyclops-64 Architecture/Software • Cyclops-64 Programming Model and System Software • Future Directions • Summary 2007/6/14 SOS11-06-2007.ppt 2 TIPs of compute power operating on Tera-bytes of data Transistor Growth in the near future Source: Keynote talk in CGO & PPoPP 03/14/07 by Jesse Fang from Intel 2007/6/14 SOS11-06-2007.ppt 3 Outline • Introduction • Multi-Core Chip Technology • IBM Cyclops-64 Architecture/Software • Programming/Compiling for Cyclops-64 • Looking Beyond Cyclops-64 • Summary 2007/6/14 SOS11-06-2007.ppt 4 Two Types of Multi-Core Architecture Trends • Type I: Glue “heavy cores” together with minor changes • Type II: Explore the parallel architecture design space and searching for most suitable chip architecture models. 2007/6/14 SOS11-06-2007.ppt 5 Multi-Core Type II • New factors to be considered –Flops are cheap! –Memory per core is small –Cache-coherence is expensive! –On-chip bandwidth can be enormous! –Examples: Cyclops-64, and others 2007/6/14 SOS11-06-2007.ppt 6 Flops are Cheap! An example to illustrate design tradeoffs: • If fed from small, local register files: 64-bit FP – 3200 GB/s, 10 pJ/op unit – < $1/Gflop (60 mW/Gflop) (drawn to a 64-bit FPU is < 1mm^2 scale) and ~= 50pJ • If fed from global on-chip memory: Can fit over 200 on a chip.
    [Show full text]
  • Virtualized Baseband Units Consolidation in Advanced Lte Networks Using Mobility- and Power-Aware Algorithms
    San Jose State University SJSU ScholarWorks Master's Projects Master's Theses and Graduate Research Fall 2017 VIRTUALIZED BASEBAND UNITS CONSOLIDATION IN ADVANCED LTE NETWORKS USING MOBILITY- AND POWER-AWARE ALGORITHMS Uladzimir Karneyenka San Jose State University Follow this and additional works at: https://scholarworks.sjsu.edu/etd_projects Part of the Computer Sciences Commons Recommended Citation Karneyenka, Uladzimir, "VIRTUALIZED BASEBAND UNITS CONSOLIDATION IN ADVANCED LTE NETWORKS USING MOBILITY- AND POWER-AWARE ALGORITHMS" (2017). Master's Projects. 571. DOI: https://doi.org/10.31979/etd.sg3g-dr33 https://scholarworks.sjsu.edu/etd_projects/571 This Master's Project is brought to you for free and open access by the Master's Theses and Graduate Research at SJSU ScholarWorks. It has been accepted for inclusion in Master's Projects by an authorized administrator of SJSU ScholarWorks. For more information, please contact [email protected]. VIRTUALIZED BASEBAND UNITS CONSOLIDATION IN ADVANCED LTE NETWORKS USING MOBILITY - AND POWER -AWARE ALGORITHMS A Writing Project Report Presented to The Faculty of the Department of Computer Science San Jose State University In Partial Fulfillment Of the Requirements for the Degree Master of Science By Uladzimir Karneyenka May 2017 © 2017 Uladzimir Karneyenka ALL RIGHTS RESERVED ABSTRACT Virtualization of baseband units in Advanced Long-Term Evolution networks and a rapid performance growth of general purpose processors naturally raise the interest in resource multiplexing. The concept of resource sharing and management between virtualized instances is not new and extensively used in data centers. We adopt some of the resource management techniques to organize virtualized baseband units on a pool of hosts and investigate the behavior of the system in order to identify features which are particularly relevant to mobile environment.
    [Show full text]
  • Simulating Linux Clusters on Linux Clusters
    Full Circle: Simulating Linux Clusters on Linux Clusters ¡ ¢ ¡ Luis Ceze , Karin Strauss , George Almasi , Patrick J. Bohrer , Jose´ R. Brunheroto , ¡ ¡ ¡ ¡ Calin Cas¸caval , Jose´ G. Castanos˜ , Derek Lieber , Xavier Martorell , ¡ ¡ ¡ Jose´ E. Moreira , Alda Sanomiya , and Eugen Schenfeld £ luisceze,kstrauss ¤ @uiuc.edu £ gheorghe,pbohrer,brunhe,cascaval,castanos,lieber, xavim,jmoreira,sanomiya,eugen ¤ @us.ibm.com ¥ Department of Computer Science, University of Illinois at Urbana-Champaign Urbana, IL 61801 ¦ IBM Thomas J. Watson Research Center Yorktown Heights, NY 10598-0218 § IBM Austin Research Laboratory Austin, TX 78758 Abstract. BGLsim is a complete system simulator for parallel machines. It is currently being used in hardware validation and software development for the BlueGene/L cellular architecture machine. BGLsim is capable of functionally simulating multiple nodes of this machine operating in parallel. It simulates in- struction execution in each node and the communication that happens between nodes. BGLsim allows us to develop, test, and run the exactly same code that will be used in the real system. Using BGLsim, we can gather data that helps us debug and enhance software (including parallel software) and evaluate hard- ware. To illustrate the capabilities of BGLsim, we describe experiments running the NAS Parallel Benchmark IS on a simulated BlueGene/L machine. BGLsim is a parallel application that runs on Linux clusters. It executes fast enough to run complete operating systems and complex MPI codes. 1 Introduction Linux clusters have revolutionized high-performance computing by delivering large amounts of compute cycles at a low price. This has enabled computing at a scale that was previously not affordable to many people.
    [Show full text]
  • Evaluating Cyclops64
    Mstack: A Lightweight Cross-Platform Benchmark for Evaluating Co-processing Technologies by Mark Pellegrini and Daniel M. Pressel ARL-MR-0683 December 2007 Approved for public release; distribution is unlimited. NOTICES Disclaimers The findings in this report are not to be construed as an official Department of the Army position unless so designated by other authorized documents. Citation of manufacturer’s or trade names does not constitute an official endorsement or approval of the use thereof. Destroy this report when it is no longer needed. Do not return it to the originator. Army Research Laboratory Aberdeen Proving Ground, MD 21005-5067 ARL-MR-0683 December 2007 Mstack: A Lightweight Cross-Platform Benchmark for Evaluating Co-processing Technologies Mark Pellegrini and Daniel M. Pressel Computational and Information Sciences Directorate, ARL Approved for public release; distribution is unlimited. REPORT DOCUMENTATION PAGE Form Approved OMB No. 0704-0188 Public reporting burden for this collection of information is estimated to average 1 hour per response, including the time for reviewing instructions, searching existing data sources, gathering and maintaining the data needed, and completing and reviewing the collection information. Send comments regarding this burden estimate or any other aspect of this collection of information, including suggestions for reducing the burden, to Department of Defense, Washington Headquarters Services, Directorate for Information Operations and Reports (0704-0188), 1215 Jefferson Davis Highway, Suite 1204, Arlington, VA 22202-4302. Respondents should be aware that notwithstanding any other provision of law, no person shall be subject to any penalty for failing to comply with a collection of information if it does not display a currently valid OMB control number.
    [Show full text]
  • Efficient Synchronization for a Large-Scale Multi-Core Chip Architecture
    EFFICIENT SYNCHRONIZATION FOR A LARGE-SCALE MULTI-CORE CHIP ARCHITECTURE by Weirong Zhu A dissertation submitted to the Faculty of the University of Delaware in partial fulfillment of the requirements for the degree of Doctor of Philosophy in Electrical and Computer Engineering Spring 2007 c 2007 Weirong Zhu All Rights Reserved UMI Number: 3267164 Copyright 2007 by Zhu, Weirong All rights reserved. UMI Microform 3267164 Copyright 2007 by ProQuest Information and Learning Company. All rights reserved. This microform edition is protected against unauthorized copying under Title 17, United States Code. ProQuest Information and Learning Company 300 North Zeeb Road P.O. Box 1346 Ann Arbor, MI 48106-1346 EFFICIENT SYNCHRONIZATION FOR A LARGE-SCALE MULTI-CORE CHIP ARCHITECTURE by Weirong Zhu Approved: Gonzalo R. Arce, Ph.D. Chairperson of the Department of Electrical and Computer Engineering Approved: Eric W. Kaler, Ph.D. Dean of the College of Engineering Approved: Carolyn A. Thoroughgood, Ph.D. Vice Provost for Research and Graduate Studies I certify that I have read this dissertation and that in my opinion it meets the academic and professional standard required by the University as a dissertation for the degree of Doctor of Philosophy. Signed: Guang R. Gao, Ph.D. Professor in charge of dissertation I certify that I have read this dissertation and that in my opinion it meets the academic and professional standard required by the University as a dissertation for the degree of Doctor of Philosophy. Signed: Fouad Kiamilev, Ph.D. Member of dissertation committee I certify that I have read this dissertation and that in my opinion it meets the academic and professional standard required by the University as a dissertation for the degree of Doctor of Philosophy.
    [Show full text]
  • Toward a Software Infrastructure for the Cyclops-64 Cellular Architecture
    Toward a Software Infrastructure for the Cyclops-64 Cellular Architecture JuandelCuvillo WeirongZhu ZiangHu GuangR.Gao Department of Electrical and Computer Engineering University of Delaware Newark, Delaware 19716, U.S.A {jcuvillo,weirong,hu,ggao}@capsl.udel.edu Abstract tem through a number of Gigabit Ethernet links. The host system provides a familiar computing environment to appli- This paper presents the initial design of the Cyclops-64 cations software developers and end users. Besides access (C64) system software infrastructure and tools under devel- (through the Ethernet links) to a common file server, each opment as a joint effort between IBM T.J. Watson Research C64 chip can be connected to a serial ATA disk drive. Center, ETI Inc. and the University of Delaware. The C64 A C64 is built out of tens of thousands of C64 pro- system is the latest version of the Cyclops cellular architec- cessing nodes. Each processing node consists of a C64 ture that consists of a large number of compute nodes each chip, external DRAM, and a small amount of external in- employs a multiprocessor-on-a-chip architecture with 160 terface logic. A C64 chip employs a multiprocessor-on-a- hardware thread units. The first version of the C64 system chip architecture containing 160 hardware thread units, half software has been developed and is now under evaluation. as many floating point units, on-chip SRAM, on-chip in- The current version of the C64 software infrastructure in- struction cache, bidirectional inter-chip routing ports, and cludes a C64 toolchain (compiler, linker, functionally accu- interface to off-chip DDR SDRAM.
    [Show full text]