+612 9045 4394
 
CHECKOUT
$7.95 Delivery per order to Australia and New Zealand
100% Australian owned
Over a hundred thousand in-stock titles ready to ship
Sci: Scalable Coherent Interface : Architecture and Software for High-Performance Compute Clusters - Hermann Hellwagner

Sci: Scalable Coherent Interface

Architecture and Software for High-Performance Compute Clusters

By: Hermann Hellwagner (Editor), Alexander Reinefeld (Editor)

Paperback Published: 20th October 1999
ISBN: 9783540666967
Number Of Pages: 494

Share This Book:

Paperback

$169.73
or 4 easy payments of $42.43 with Learn more
Ships in 10 to 15 business days

Earn 339 Qantas Points
on this Book

Scalable Coherent Interface (SCI) is an innovative interconnect standard (ANSI/IEEE Std 1596-1992) addressing the high-performance computing and networking domain. This book describes in depth one specific application of SCI: its use as a high-speed interconnection network (often called a system area network, SAN) for compute clusters built from commodity workstation nodes. The editors and authors, coming from both academia and industry, have been instrumental in the SCI standardization process, the development and deployment of SCI adapter cards, switches, fully integrated clusters, and software systems, and are closely involved in various research projects on this important interconnect. This thoroughly cross-reviewed state-of-the-art survey covers the complete hardware/software spectrum of SCI clusters, from the major concepts of SCI, through SCI hardware, networking, and low-level software issues, various programming models and environments, up to tools and application experiences.

SCI and Competitive Interconnects for Cluster Computing
The SCI Standard and Applications of SCIp. 3
Introductionp. 3
SCI Overviewp. 4
Backgroundp. 4
Goalsp. 4
Conceptsp. 6
Discussionp. 11
The SCI Standard and Some Extensionsp. 11
Logical Layerp. 12
Cache Coherence Layerp. 19
Extensionsp. 22
Applications of SCIp. 23
System Area Network for Clustersp. 23
Memory Interconnect for Cache-Coherent Multiprocessorsp. 26
I/O Subsystem Interconnectp. 30
Large-Scale Data Acquisition Systemp. 31
Related Communication Networks and Conceptsp. 31
Concluding Remarksp. 34
A Comparison of Three Gigabit Technologies: SCI, Myrinet and SGI/Cray T3Dp. 39
Introductionp. 39
Levels of Comparisonp. 40
Direct Depositp. 41
Message Passing (MPI/PVM)p. 42
Protocol Emulation (TCP/IP)p. 44
Gigabit Network Technologiesp. 45
The Intel 80686 Hardware Platformp. 46
Myricom Myrinet Technologyp. 47
Dolphin PCI-SCI Technologyp. 48
The SGI/Cray T3D - A Reference Pointp. 48
ATM: QoS - But Still Short of a Gigabit/sp. 50
Gigabit Ethernet - An Outlookp. 50
Transfer Modesp. 51
Overviewp. 51
"Native" and "Alternate" Transfer Modes in the Three Architecturesp. 54
Performance Evaluationp. 56
Performance of Local Memory Copyp. 58
Performance of Direct Transfers to Remote Memoryp. 58
Performance of MPI/PVM Transfersp. 61
Performance of TCP/IP Transfersp. 64
Discussion and Comparisonp. 65
Summaryp. 67
SCI Hardware
Dolphin SCI Adapter Cardsp. 71
Introductionp. 71
Overview of the Adapter Cardsp. 71
Operating Modes of the SCI Cardsp. 73
SCI Requesterp. 74
Address Mappingp. 74
SCI Transaction Handlingp. 75
SCI Packet Requesterp. 77
SCI Responderp. 78
Mailboxp. 79
Access Protectionp. 79
Atomic Accessp. 79
Host Bridge Capabilitiesp. 80
DMA Transfersp. 80
DMA Transfers on the SBus Cardp. 80
DMA Transfers on the PCI Cardp. 80
Interrupterp. 81
Concurrency Issuesp. 81
Write Assemblyp. 81
Efficient Store Barrierp. 81
Performancep. 82
Applications and Topologiesp. 82
SAN Interface Adapterp. 83
Remote I/O Connection and Data Acquisitionp. 83
Switches and Topologiesp. 83
Cluster Softwarep. 85
The TUM PCI/SCI Adapterp. 89
Introductionp. 89
The PCI/SCI Adapter Architecturep. 90
SCI Packet Encoding and Decodingp. 92
Overview of Packet Processingp. 92
Choosing the Technologyp. 92
Internal Structure of the FPGAp. 93
Structure of the Packet Manag er as a Microcode Sequencerp. 95
Microcode Examplesp. 97
Benefits of the Micro Sequencerp. 98
The SCI Unitp. 99
Preliminary Results for the PCI/SCI Adapterp. 99
Related Workp. 100
Conclusionp. 100
Interconnection Networks with SCI
Low-Level SCI Protocols and Their Application to Flexible Switchesp. 105
Introductionp. 105
Data Format of SCI Packetsp. 105
Flow Controlp. 107
Flow Control in Ringsp. 107
Packet Sequence in SCIp. 108
Determination of State Transitionsp. 109
Bandwidth Multiplexingp. 110
Bandwidth Management in One Ringp. 110
Idle Symbolsp. 112
Time-Out Determinationp. 113
Network Interfacep. 113
Requirementsp. 114
Productsp. 114
Routersp. 115
Requirementsp. 115
Products and Challengesp. 116
Flexible Routerp. 117
Strip-off Decisionp. 118
Routing Decision and Topologyp. 119
Rule-Based Routingp. 120
Conclusion and Outlookp. 121
SCI Rings, Switches, and Networks for Data Acquisition Systemsp. 125
Introductionp. 125
SCI-based Data Acquisition Systemsp. 126
SCINET Test Bedsp. 127
Measurement Resultsp. 129
SCI Switchesp. 134
Efficient Use of SCI Switchesp. 136
Multistage SCI Networksp. 139
Simulation Resultsp. 141
Summary and Conclusionsp. 146
Scalability of SCI Ringletsp. 151
Do SCI Ringlets Scale in Number of Nodes?p. 151
Ringlet Bandwidth Modelp. 152
Transaction Formatsp. 152
Packet Generationp. 155
Address Distributionp. 155
Localityp. 156
Bypass Ratep. 157
Echo Packet Ratep. 158
Output Link Utilization Factorp. 160
Scalability Evaluationp. 160
Common Assumptionsp. 161
Uniform Ringlet Trafficp. 162
Non-uniform Ringlet Trafficp. 162
Changing Packet Lengthsp. 163
Discussionp. 163
Conclusionp. 165
Affordable Scalability Using Multi-Cubesp. 167
Introductionp. 167
Interconnect Overviewp. 168
Methodologyp. 168
Analysisp. 170
"Hot-Link" Analysisp. 170
"Hot-B-Link" Analysisp. 171
Resultsp. 172
Conclusionsp. 174
Device Driver Software and Low-Level APIs
Interfacing SCI Device Drivers to Linuxp. 179
Introductionp. 179
Layers of Functionalityp. 180
Address Spacesp. 180
Levels of Hardware Abstractionp. 180
Resource Managementp. 182
Virtual Mappingp. 183
Robustnessp. 184
Why Linux?p. 185
Interfaces of the Driverp. 186
Hardwarep. 186
Linuxp. 187
User Processesp. 188
SCI Drivers on Other Nodesp. 188
Conclusionsp. 189
SCI Physical Layer APIp. 191
Introductionp. 191
Scope of the Standardp. 192
SCI Physical Layer API Architecture and Featuresp. 193
Exception Handlingp. 195
Endiannessp. 195
Supported Data Typesp. 196
Miscellaneous Proceduresp. 196
Address Translation Modelp. 197
Global Object Identifierp. 199
SCI Global Address Resolutionp. 200
Shared Memory Transactionsp. 200
Packet Transactionsp. 202
Block Transactionsp. 202
Message Passing Transactionsp. 203
Cache Transactionsp. 204
Conclusionsp. 205
Message Passing Libraries
SCI Sockets Libraryp. 209
Introductionp. 209
Rationalep. 209
Overviewp. 210
Features and Designp. 210
Featuresp. 210
Componentsp. 211
Communication via the SSLibp. 212
Connection Setupp. 214
Handling Special System Callsp. 216
Other Calls Intercepted and Handled by the SSLibp. 218
Out of Band Datap. 218
Implementation Aspectsp. 218
Communication Among Componentsp. 218
SSLib Layersp. 219
Choice of Most Efficient Communication Mechanismp. 220
SSLib Implementationsp. 221
Control Transfersp. 221
Functional Tests and Performancep. 222
Related Workp. 224
Conclusionsp. 227
TCP/IP over SCI under Linuxp. 231
Introductionp. 231
SCIP Structurep. 232
Packet Driver Interfacep. 232
Hardware Address Resolutionp. 232
Other Implementation Issuesp. 233
Performancep. 234
Configurationp. 234
Latencyp. 234
Throughputp. 235
Conclusionp. 237
PVM for SCI Clustersp. 239
Overviewp. 239
Parallel Virtual Machinep. 239
PVM Implementationsp. 240
Models for Zero-Memory-Copy Data Transferp. 241
SCI Communication Modelp. 242
PVM-SCIp. 243
System Architecturep. 243
Supporting Multiple Interconnectsp. 245
Reducing Memory Copiesp. 245
Ring Buffer Managementp. 246
Performance Resultsp. 247
Conclusionsp. 247
ScaMPI - Design and Implementationp. 249
Introductionp. 249
Scali Systemsp. 249
The SCI Memory Modelp. 250
Coordinating Use of Shared Locationsp. 251
Ensuring Safe Data Transport in SCI - Checkpointingp. 252
Shared Address Space Programming without the Drawbacksp. 252
ScaMPI Design Goalsp. 253
ScaMPI Implementationp. 254
Fault Tolerancep. 254
User Friendlinessp. 256
Third Party Softwarep. 256
Performance Resultsp. 257
Barrierp. 258
All-to-All Communicationp. 259
Conclusionsp. 260
Shared Memory Programming Models and Runtime Mechanisms
Shared Memory vs Message Passing on SCI: A CaseStudy Using Split-Cp. 267
Introductionp. 267
Introduction to Split-Cp. 268
Introduction to Active Messagesp. 269
Message-Passing Implementationp. 269
Active Messages on Top of SCIp. 269
Split-C on Top of Active Messagesp. 272
Shared Memory Implementationp. 273
Split-C on Top of SCIp. 273
Experimental Evaluationp. 274
Micro-benchmarksp. 274
Table of Contents provided by Publisher. All Rights Reserved.

ISBN: 9783540666967
ISBN-10: 3540666966
Series: Lecture Notes in Computer Science
Audience: General
Format: Paperback
Language: English
Number Of Pages: 494
Published: 20th October 1999
Publisher: Springer-Verlag Berlin and Heidelberg Gmbh & Co. Kg
Country of Publication: DE
Dimensions (cm): 23.39 x 15.6  x 2.64
Weight (kg): 0.72

Earn 339 Qantas Points
on this Book