The Intel Paragon
Total Page：16
File Type：pdf, Size：1020Kb
Operating system and network support for high-performance computing Item Type text; Dissertation-Reproduction (electronic) Authors Guedes Neto, Dorgival Olavo Publisher The University of Arizona. Rights Copyright © is held by the author. Digital access to this material is made possible by the University Libraries, University of Arizona. Further transmission, reproduction or presentation (such as public display or performance) of protected items is prohibited except with permission of the author. Download date 30/09/2021 20:24:35 Link to Item http://hdl.handle.net/10150/298757 INFORMATION TO USERS This manuscript has been reproduced from the microfilm master. UMI films the text directly from the original or copy submitted. Thus, some thesis and dissertation copies are in typewriter face, while others may be from any type of computer printer. The quality of this reproduction is dependent upon the quality of the copy submitted. Broken or indistinct print, colored or poor quality illustrations and photographs, print bleedthrough, substandard margins, and improper alignment can adversely affect reproduction. In the unlikely event that the author did not send UMI a complete manuscript and there are missing pages, these will be noted. Also, if unauthorized copyright material had to be removed, a note will indicate the deletion. Oversize materials (e.g., maps, drawings, charts) are reproduced by sectioning the original, beginning at the upper left-hand comer and continuing from left to right in equal sections with small overlaps. Each original is also photographed in one exposure and is included in reduced form at the back of the book. Photographs included in the original manuscript have been reproduced xerographically in this copy. Higher quality 6" x 9" black and white photographic prints are available for any photographs or illustrations appearing in this copy for an additional charge. Contact UMI directly to order. Bell & Howell Information and Learning 300 North Zeeb Road, Ann Arbor, Ml 48106-1346 USA 800-521-0600 OPERATING SYSTEM AND NETWORK SUPPORT FOR HIGH-PERFORMANCE COMPUTING by Dorgival Olavo Guedes Neto Copyright © Dorgival Olavo Guedes Neto 1999 A Dissertation Submitted to the Faculty of the DEPARTMENT OF COMPUTER SCIENCE In Partial Fulfillment of the Requirements For the Degree of DOCTOR OF PHILOSOPHY In the Graduate College THE UNIVERSITY OF ARIZONA 19 9 9 UMI Number: 9946820 Copyright 1999 by Guedes Neto, Dorgival Olavo All rights reserved. UMI Microfonn 9946820 Copyright 1999, by UMI Company. All rights reserved. This microfonn edition is protected against unauthorized copying under Title 17, United States Code. UMI 300 North Zeeb Road Ann Arbor, MI 48103 2 THE UNIVERSITY OF ARIZONA ® GRADUATE COLLEGE As members of the Final Examination Conmittee, we certify that we have read the dissertation prepared by Dorgival Olavo Guedes Neto entitled OPERATING SYSTEM AND NETWORK SUPPORT FOR HIGH-PERFORMANCE COMPUTING and reconmend that it be accepted as fulfilling the dissertation requirement for the Degree of Doctor of Philosophy s-U-ij'21. eterson DatTl ^ ^ohn H. Hartman Date Richard D. SchlichtDate Date Date Final approval and acceptance of this dissertation is contingent upon the candidate's submission of the final copy of the dissertation to the Graduate College. I hereby certify that I have read this dissertation prepared under my direction and reconmend that it be accepted as fulfilling the dissertation requirement. Dissertation director Larry L. Peterson Date 3 STATEMENT BY AUTHOR This dissertation has been submitted in partial fulfillment of requirements for an advanced degree at The University of Arizona and is deposited in the University Libreiry to be made available to borrowers under rules of the Library. Brief quotations from this dissertation are allowable without special permission, pro vided that accurate acknowledgment of source is made. Requests for permission for extended quotation from or reproduction of this manuscript in whole or in part may be granted by the copyright holder. SIGNED: 4 ACKNOWLEDGEMENTS The road to a Ph.D. is always an adventurous one, and I have been greatly honored to have Ljirry Peterson as my jidvisor along the way. Many were the times when I came to his office feeling like in a dead end, and left with new confidence and new paths to explore. He was always there to hear ajiy ideas I had, to help turn bad ones into good, and to make good ones even better, with insights often beyond my own. Working with him has been a reward in itself. I also would like to thank the other members of my committee, John Hartman, Rick Schlichting, and Greg Andrews (who served in it for a while), for their support and encouragement. I could not have made it without the help of the lab staff of the Computer Science department. In particular, I would like to thank John Cropper, for his patience anwering tons of questions and going over old backups to help me fix many of my mistakes, £tnd Phil Kaslo, for being always ready to help me with the Paragon, managing to keep it running long enough for me to get all the numbers I needed. Also, the administrative stjiff has always been there to help, specially Margaret Newman, Cara Wallace and Wendy Swartz. Many fellow students also helped, in different ways. Among them, the following deserve special mention: David Mosberger provided the Scout thread package he wrote for the Alpha, which I ported to the Paragon; Lawrence Brakmo helped me with x-sim and offered veduable comments on rate-bcised traffic shaping; and Robert Muth helped a lot during the preparation of the presentation, providing invaluable comments on earlier versions of the slides. Sometimes it is hard to identify the moment when we started a long journey such as a doctorate. In my case, I remember it exactly. For that I thank my M.S. advisor at the UFMG, Osvaldo S. F. de Carvalho. Neither of us had any idea about it then, but when he asked me to lead a discussion about one of the first i-kernel papers ([HP91]), that was the start. Above all else, my wife, Daniella, deserves special thanks for her understanding, love and support in all moments. She unselfishly put her career on hold to keep me company diu-ing all this time, and for this I will be forever grateful. I could never have completed his work without her by my side, and I share this achievement with her. Finally, I would like to thank my parents for their education and support. During my doctorate I was particdly supported by the Brazilian Science and Technol ogy Council (Conselho Brasileiro de Desenvolvimento Cientffico e Tecnologico, CNPq), scholarship no. 200861/93-0. This work vas also supported in part by Darpa (California Institute of Technology subcontract PC159228). 5 To Daniella, for her love and support, adways. And to my grandfather, the first Dorgival Guedes (in memorian). ... "Benga, V6!" 6 TABLE OF CONTENTS LIST OF FIGURES 10 LIST OF TABLES 12 ABSTRACT 13 CHAPTER 1: INTRODUCTION 14 1.1 The Parallel I/O Bottleneck 16 1.2 Moving Data Out of the MPP 17 1.2.1 The Need to Share Information 17 1.2.2 Meta-Computing 17 1.2.3 New Network Technologies 18 1.2.4 Network Attached Peripherals 19 1.2.5 Workstation Clusters 19 1.2.6 Standeird Parallel I/O Interfaces 19 1.2.7 The Web 20 1.3 Problems With MPP Network Subsystems 20 1.4 Thesis Statement £ind Contributions 21 1.5 Overview of This Dissertation 22 CHAPTER 2: PARALLEL I/O ISSUES 23 2.1 The ParjiUel I/O Problem 23 2.2 I/O Nodes in Multiprocessor Architectures 24 2.3 Parallel File Systems 27 2.3.1 Operating System Organization 27 2.3.2 File Layout 30 2.3.3 Access Interfaces 30 2.3.4 Strided Accesses and Collective I/O 31 TABLE OF CONTENTS - Continued 7 2.3.5 Current Parallel File Systems 33 2.3.6 Communication Patterns 34 2.4 Impact on Network I/O 35 2.4.1 Network I/O Nodes 36 2.4.2 Network OS Subsystems 37 2.4.3 Communication Patterns Over the External Network 39 2.4.4 Protocol Requirements 41 2.5 Concluding Remarks 42 CHAPTER 3: DISTRIBUTED PROTOCOL PROCESSING 44 3.1 A Case Study: The Intel Paragon 44 3.1.1 Architecture 45 3.1.2 Operating System Structure 46 3.1.3 Accessible Interfaces 47 3.1.4 The network subsystem 47 3.1.5 The Target Machines 50 3.2 Performance of the Current System 51 3.2.1 OSF/1 Protocol Server 51 3.2.2 Inter-node Communication 53 3.2.3 HIPPI Interface 54 3.3 Distributed Protocol Stacks 56 3.3.1 User-Level Protocols 56 3.3.2 ULP in the Paragon 57 3.3.3 Raw HIPPI Implementation 59 3.3.4 Raw HIPPI Performance Results 60 3.3.5 NX Implementation 61 3.3.6 NX Performance Results 62 3.4 Comparison of the Different Solutions 63 3.5 Concluding Remarks 64 CHAPTER 4: PROTOCOL ISSUES AFFECTING PARALLEL I/O 66 4.1 Window-Based Flow Control 66 4.2 Congestion 67 TABLE OF CONTENTS - Contiaued 8 4.3 Connection Start-up 68 4.3.1 Setup Haudshake 68 4.3.2 TCP Slow Start 69 4.3.3 Proposals to Improve Slow Start 70 4.4 Congestion Avoidance Phase 71 4.4.1 Original TCP Congestion Control 71 4.4.2 Proposals to improve TCP Congestion Control 72 4.5 Problems of TCP for High Performance Computing 74 4.5.1 Explicit Delays 74 4.5.2 Timer Granularity 76 4.5.3 Packet TVains 77 4.6 Cooperative Sessions 81 4.7 Concluding Remarks 84 CHAPTER 5: COOPERATIVE RATE-BASED TRAFFIC SHAPING 85 5.1 Implementation 86 5.1.1 Rate Estimation 87 5.1.2 Combination of Multiple Connections 89 5.1.3 Connection Scheduling 91 5.1.4 Fine-GraJned Timers 91 5.1.5 Operation of the Rate Controller 92 5.2 Simulation Models 93 5.3 Performance Results 96 5.3.1 Single Switch Case: Read 97 5.3.2 Single Switch Case: Write 101