Sun's High Performance Computing Reference Architecture 
by Torben Kling-Petersen, Börje Lindh, and Ola Tørudbakken
April 2009
This Sun BluePrints article is designed to provide IT managers with a grounding in the basic products and technologies that comprise Suns' HPC Reference Architecture. Key components are highlighted along with typical uses, issues, and architectures of HPC systems.
- "Sun's HPC Reference Architecture" gives brief background on HPC, clustering, and introduces Sun's HPC Reference Architecture
- "Compute Systems" provides an overview of compute node for clusters and highlights the diverse range of compute nodes available from Sun.
- "Interconnects and High-Performance Networks" provides background on Ethernet networks and InfiniBand interconnects, and introduces Sun Datacenter Switches for DDR and QDR InfiniBand fabrics
- "File Systems and Archive Solutions for HPC" outlines the methods for feeding data to the cluster as well serving home directories and providing solutions for archiving data
- "HPC Software, Resource Management, and System Management" describes Sun HPC Software along with resource and system management software
Contents
- Introduction
- Sun's HPC Reference Architecture
- Improved efficiency through HPC
- Ubiquitous clustering technology
- Sun's end-to-end architecture for HPC
- Proven reference designs for HPC
- Compute Systems
- General requirements
- x86/x64 servers for HPC
- Chip multithreaded (CMT) SPARC® systems for HPC
- Interconnects and High-Performance Networks
- Ethernet
- InfiniBand
- Sun Datacenter Switches for InfiniBand infrastructure
- File Systems and Archive Solutions for HPC
- Network File System (NFS)
- Lustre parallel file system
- Sun Storage Cluster and the Lustre parallel file system
- Solari ZFS and Sun Storage 7000 Unified Storage Systems
- Sun Storage and Archive Solution for HPC
- HPC Software, Resource Management, and System Management
- Suns HPC Software for Linux and the Solaris OS
- Resource management with Sun Grid Engine software
- System management
- Summary
•Torben Kling-Petersen, Ph.D. — Senior Technical Specialist, HPC, Lustre Group
Torben Kling-Petersen has worked with high performance computing in one form or another since 1994 and is currently working as a Senior Technical Specialist for HPC in Sun's Global Systems Practice. Over the years, he has worked in a number of capacities such as lead architect for enterprise datacenter infrastructure, technical research lead and product specialist for high-end visualization to mention a few. In his present capacity, Torben works in a global role providing technical evangelism and solution architectures on petaflop-scale HPC projects.
•Börje Lindh — Senior Systems Engineer
Borje Lindh has an M.Sc. in Chemical Engineering but has been working with Computers since 1987. He has been with Sun Microsystems since 1994, working in a number of different roles and has been involved with High Performance computing since 1997. He has written two Sun Blueprints on compute clusters and a some magazine articles mainly on processor architecture.
•Ola Tørudbakken — Distinguished Engineer, Scalable Systems Group
Ola Tørudbakken has an M.Sc. degree from the University of Oslo, Department of Informatics in 1994, and has since then been working on High Performance Interconnects and Server Systems. He has been with Sun Microsystems since 2000, and his current responsibilities includes supervision and architectural definition of Fabric Products. He has published several papers in the field of interconnection networks and has participated in several standardization activities. He currently holds three US Patents, and has more than 30 US patents pending.
The authors would like to recognize Eric Liefeld, an independent technical specialist and writer for his assistance with this Sun BluePrints article. Eric is a former Sun Systems Engineer and a frequent contributor to Sun Microsystems technical documents.
| Rate this blueprint (Log In to vote.) | |
|---|---|
| Choices | Your Vote |
Great |
|
Good |
|
Fair |
|
Poor |
|
Sign up or Log in to add a comment or watch this page.