Solid State Drives in HPC: Reducing the I/O Bottleneck 
by Lawrence McIntosh and Michael Burke
June 2009
This Sun BluePrints™ article focuses on a comparison between traditional hard disk drives (HDDs) and the newer solid state drive (SSD) technology in high-performance computing (HPC) applications. SSD devices can help correct the imbalance between processor and storage speed while also reducing energy usage and environmental impact. This comparison was performed using two approaches:
- Application-based benchmarking was performed using the ABAQUS, NASTRAN, and ANSYS finite-element analysis (FEA) applications, in order to evaluate the effect of SSD technology in realistic HPC applications. These applications are commonly used to benchmark HPC systems.
- Benchmark testing of storage performance using the Lustre™ parallel file system and the popular IOZone benchmark application was performed, in order to evaluate large sequential I/O operations typical for the Lustre file system employed as a compute cluster data cache. These tests were performed using three system configurations:
- A baseline test using the Lustre file system with a single HDD-based Object Storage Server (OSS)
- A Lustre file system configuration using a single SSD-based OSS similar to the baseline test
- A comparison test using the Lustre file system and two SSD-based OSSs in parallel
The results of these tests demonstrate the potential for significant benefits in the use of SSD devices for HPC applications with large I/O components.
The results of these tests demonstrate the potential for significant benefits in the use of SSD devices for HPC applications with large I/O components.
Contents
- Introduction
- Motivation
- SSD technology review
- Single system application performance
- The ABAQUS benchmark application
- Hardware configuration
- Software configuration
- The NASTRAN benchmark application
- Hardware configuration
- Software configuration
- The ANSYS benchmark application
- Hardware configuration
- Software configuration
- Summary for single system application performance
- The ABAQUS benchmark application
- SSD usage with the Lustre parallel file system
- Lustre file system design
- IOZone file system testing
- Hardware configuration
- Software configuration
- Summary for SSD usage with the Lustre parallel file system
- Future directions
- Conclusion
- Appendix: Benchmark descriptions and parameters
- ABAQUS standard benchmark test cases
- Hardware configuration
- Software configuration
- NASTRAN benchmark test cases
- Hardware configuration
- Software configuration
- ANSYS 12.0 (prel. 7) with ANSYS 11.0 distributed benchmarks
- Hardware configuration
- Software configuration
- ABAQUS standard benchmark test cases
- About the authors
- References
- Ordering Sun Documents
- Accessing Sun Documentation Online
Larry McIntosh is a Principal Systems Engineer at Sun Microsystems and works within Sun's Systems Engineering Solutions Group. He is responsible for designing and implementing high performance computing technologies at Sun's largest customers. Larry has 35 years of experience in the computer, communications, and storage industries and has been a software developer and consultant in the commercial, government, education and research sectors as well as a computer science college professor. Larry's recent work has included the deployment of the Ranger system servicing the National Science Foundation and Researchers at the Texas Advanced Computer Center (TACC) in Austin, Texas.
Michael Burke obtained his Ph.D. from Stanford University. Since then he has spent over 35 years in the development and application of MCAE software. He was the principal developer of the MARC code now owned by MSC/Nastran. Following the SS Challenger disaster he developed FANTASTIC (Failure Analysis Thermal and Structural Integrated Code) for NASA and its suppliers/contractors for the analysis of rocket (nozzles) More recently he has been involved with the benchmarking of state of the art HPC platforms using the more prominent commercial ISV MCAE/CFD/CRASH and other scientific applications He has performed this benchmarking for Fujitsu and Hewlett Packard, and is currently in the Strategic Applications Engineering group at Sun Microsystems.
| 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.