SCI and Competitive Interconnects for Cluster Computing / Part I: |
The SCI Standard and Applications of SCI / Hermann Hellwagner1: |
Introduction / 1.1: |
SCI Overview / 1.2: |
Background / 1.2.1: |
Goals / 1.2.2: |
Concepts / 1.2.3: |
Discussion / 1.2.4: |
The SCI Standard and Some Extensions / 1.3: |
Logical Layer / 1.3.1: |
Cache Coherence Layer / 1.3.2: |
Extensions / 1.3.3: |
Applications of SCI / 1.4: |
System Area Network for Clusters / 1.4.1: |
Memory Interconnect for Cache-Coherent Multiprocessors / 1.4.2: |
I/O Subsystem Interconnect / 1.4.3: |
Large-Scale Data Acquisition System / 1.4.4: |
Related Communication Networks and Concepts / 1.5: |
Concluding Remarks / 1.6: |
A Comparison of Three Gigabit Technologies: SCI, Myrinet and SGI/Cray T3D / Christian Kurmann ; Thomas Stricker2: |
Levels of Comparison / 2.1: |
Direct Deposit / 2.2.1: |
Message Passing (MPI/PVM) / 2.2.2: |
Protocol Emulation (TCP/IP) / 2.2.3: |
Gigabit Network Technologies / 2.3: |
The Intel 80686 Hardware Platform / 2.3.1: |
Myricom Myrinet Technology / 2.3.2: |
Dolphin PCI-SCI Technology / 2.3.3: |
The SGI/Cray T3D - A Reference Point / 2.3.4: |
ATM: QoS - But Still Short of a Gigabit/s / 2.3.5: |
Gigabit Ethernet - An Outlook / 2.3.6: |
Transfer Modes / 2.4: |
Overview / 2.4.1: |
"Native" and "Alternate" Transfer Modes in the Three Architectures / 2.4.2: |
Performance Evaluation / 2.5: |
Performance of Local Memory Copy / 2.5.1: |
Performance of Direct Transfers to Remote Memory / 2.5.2: |
Performance of MPI/PVM Transfers / 2.5.3: |
Performance of TCP/IP Transfers / 2.5.4: |
Discussion and Comparison / 2.5.5: |
Summary / 2.6: |
SCI Hardware / Part II: |
Dolphin SCI Adapter Cards / Marius Christian Liaaen ; Hugo Kohmann3: |
Overview of the Adapter Cards / 3.1: |
Operating Modes of the SCI Cards / 3.3: |
SCI Requester / 3.4: |
Address Mapping / 3.4.1: |
SCI Transaction Handling / 3.4.2: |
SCI Packet Requester / 3.4.3: |
SCI Responder / 3.5: |
Mailbox / 3.5.1: |
Access Protection / 3.5.2: |
Atomic Access / 3.5.3: |
Host Bridge Capabilities / 3.5.4: |
DMA Transfers / 3.6: |
DMA Transfers on the SBus Card / 3.6.1: |
DMA Transfers on the PCI Card / 3.6.2: |
Interrupter / 3.7: |
Concurrency Issues / 3.8: |
Write Assembly / 3.8.1: |
Efficient Store Barrier / 3.8.2: |
Performance / 3.9: |
Applications and Topologies / 3.10: |
SAN Interface Adapter / 3.10.1: |
Remote I/O Connection and Data Acquisition / 3.10.2: |
Switches and Topologies / 3.10.3: |
Cluster Software / 3.11: |
The TUM PCI/SCI Adapter / Georg Acher ; Wolfgang Karl ; Markus Leberecht4: |
The PCI/SCI Adapter Architecture / 4.1: |
SCI Packet Encoding and Decoding / 4.3: |
Overview of Packet Processing / 4.3.1: |
Choosing the Technology / 4.3.2: |
Internal Structure of the FPGA / 4.3.3: |
Structure of the Packet Manag er as a Microcode Sequencer / 4.3.4: |
Microcode Examples / 4.3.5: |
Benefits of the Micro Sequencer / 4.3.6: |
The SCI Unit / 4.4: |
Preliminary Results for the PCI/SCI Adapter / 4.5: |
Related Work / 4.6: |
Conclusion / 4.7: |
Interconnection Networks with SCI / Part III: |
Low-Level SCI Protocols and Their Application to Flexible Switches / Andreas C. Döring ; Wolfgang Obelöer ; Gunther Lustig ; Erik Maehle5: |
Data Format of SCI Packets / 5.1: |
Flow Control / 5.3: |
Flow Control in Rings / 5.3.1: |
Packet Sequence in SCI / 5.3.2: |
Determination of State Transitions / 5.3.3: |
Bandwidth Multiplexing / 5.4: |
Bandwidth Management in One Ring / 5.4.1: |
Idle Symbols / 5.4.2: |
Time-Out Determination / 5.4.3: |
Network Interface / 5.5: |
Requirements / 5.5.1: |
Products / 5.5.2: |
Routers / 5.6: |
Products and Challenges / 5.6.1: |
Flexible Router / 5.6.3: |
Strip-off Decision / 5.6.4: |
Routing Decision and Topology / 5.6.5: |
Rule-Based Routing / 5.7: |
Conclusion and Outlook / 5.8: |
SCI Rings, Switches, and Networks for Data Acquisition Systems / Harald Richter ; Richard Kleber ; Matthias Ohlenroth6: |
SCI-based Data Acquisition Systems / 6.1: |
SCINET Test Beds / 6.3: |
Measurement Results / 6.4: |
SCI Switches / 6.5: |
Efficient Use of SCI Switches / 6.6: |
Multistage SCI Networks / 6.7: |
Simulation Results / 6.8: |
Summary and Conclusions / 6.9: |
Scalability of SCI Ringlets / Geir Horn7: |
Do SCI Ringlets Scale in Number of Nodes? / 7.1: |
Ringlet Bandwidth Model / 7.2: |
Transaction Formats / 7.2.1: |
Packet Generation / 7.2.2: |
Address Distribution / 7.2.3: |
Locality / 7.2.4: |
Bypass Rate / 7.2.5: |
Echo Packet Rate / 7.2.6: |
Output Link Utilization Factor / 7.2.7: |
Scalability Evaluation / 7.3: |
Common Assumptions / 7.3.1: |
Uniform Ringlet Traffic / 7.3.2: |
Non-uniform Ringlet Traffic / 7.3.3: |
Changing Packet Lengths / 7.3.4: |
Affordable Scalability Using Multi-Cubes / HÃ¥kon Bugge ; Knut Omang7.4: |
Interconnect Overview / 8.1: |
Methodology / 8.3: |
Analysis / 8.4: |
"Hot-Link" Analysis / 8.4.1: |
"Hot-B-Link" Analysis / 8.4.2: |
Results / 8.5: |
Conclusions / 8.6: |
Device Driver Software and Low-Level APIs / Part IV: |
Interfacing SCI Device Drivers to Linux / Roger Butenuth ; Hans-Ulrich Heiss9: |
Layers of Functionality / 9.1: |
Address Spaces / 9.2.1: |
Levels of Hardware Abstraction / 9.2.2: |
Resource Management / 9.2.3: |
Virtual Mapping / 9.2.4: |
Robustness / 9.2.5: |
Why Linux? / 9.3: |
Interfaces of the Driver / 9.4: |
Hardware / 9.4.1: |
Linux / 9.4.2: |
User Processes / 9.4.3: |
SCI Drivers on Other Nodes / 9.4.4: |
SCI Physical Layer API / Volker Lindenstruth ; David B. Gustavson9.5: |
Scope of the Standard / 10.1: |
SCI Physical Layer API Architecture and Features / 10.2: |
Exception Handling / 10.2.1: |
Endianness / 10.2.2: |
Supported Data Types / 10.3: |
Miscellaneous Procedures / 10.4: |
Address Translation Model / 10.5: |
Global Object Identifier / 10.5.1: |
SCI Global Address Resolution / 10.5.2: |
Shared Memory Transactions / 10.6: |
Packet Transactions / 10.7: |
Block Transactions / 10.8: |
Message Passing Transactions / 10.9: |
Cache Transactions / 10.10: |
Message Passing Libraries / 10.11: |
SCI Sockets Library / Josef Weidendorfer11: |
Rationale / 11.1: |
Features and Design / 11.1.2: |
Features / 11.2.1: |
Components / 11.2.2: |
Communication via the SSLib / 11.2.3: |
Connection Setup / 11.2.4: |
Handling Special System Calls / 11.2.5: |
Other Calls Intercepted and Handled by the SSLib / 11.2.6: |
Out of Band Data / 11.2.7: |
Implementation Aspects / 11.3: |
Communication Among Components / 11.3.1: |
SSLib Layers / 11.3.2: |
Choice of Most Efficient Communication Mechanism / 11.3.3: |
SSLib Implementations / 11.3.4: |
Control Transfers / 11.3.5: |
Functional Tests and Performance / 11.4: |
TCP/IP over SCI under Linux / Hüseyin Taskin11.5: |
SCIP Structure / 12.1: |
Packet Driver Interface / 12.2.1: |
Hardware Address Resolution / 12.2.2: |
Other Implementation Issues / 12.2.3: |
Configuration / 12.3: |
Latency / 12.3.2: |
Throughput / 12.3.3: |
PVM for SCI Clusters / Markus Fischer ; Alexander Reinefeld12.4: |
Parallel Virtual Machine / 13.1: |
PVM Implementations / 13.2.1: |
Models for Zero-Memory-Copy Data Transfer / 13.2.2: |
SCI Communication Model / 13.3: |
PVM-SCI / 13.4: |
System Architecture / 13.4.1: |
Supporting Multiple Interconnects / 13.4.2: |
Reducing Memory Copies / 13.4.3: |
Ring Buffer Management / 13.4.4: |
Performance Results / 13.4.5: |
ScaMPI - Design and Implementation / L.P. Huse ; K. Omang ; H. Bugge ; H. Ry ; A.T. Haugsdal ; E. Rustad13.5: |
Scali Systems / 14.1: |
The SCI Memory Model / 14.3: |
Coordinating Use of Shared Locations / 14.3.1: |
Ensuring Safe Data Transport in SCI - Checkpointing / 14.3.2: |
Shared Address Space Programming without the Drawbacks / 14.3.3: |
ScaMPI Design Goals / 14.4: |
ScaMPI Implementation / 14.5: |
Fault Tolerance / 14.5.1: |
User Friendliness / 14.5.2: |
Third Party Software / 14.5.3: |
Barrier / 14.6: |
All-to-All Communication / 14.6.2: |
Shared Memory Programming Models and Runtime Mechanisms / 14.7: |
Shared Memory vs Message Passing on SCI: A CaseStudy Using Split-C / Max Ibel ; Michael Schmitt ; Klaus Schauser ; Anurag Acharya15: |
Introduction to Split-C / 15.1: |
Introduction to Active Messages / 15.1.2: |
Message-Passing Implementation / 15.2: |
Active Messages on Top of SCI / 15.2.1: |
Split-C on Top of Active Messages / 15.2.2: |
Shared Memory Implementation / 15.3: |
Split-C on Top of SCI / 15.3.1: |
Experimental Evaluation / 15.4: |
Micro-benchmarks / 15.4.1: |
SCI and Competitive Interconnects for Cluster Computing / Part I: |
The SCI Standard and Applications of SCI / Hermann Hellwagner1: |
Introduction / 1.1: |