WARNING: This is the _old_ Lustre wiki, and it is in the process of being retired. The information found here is all likely to be out of date. Please search the new wiki for more up to date information.
Metadata Performance Project: Difference between revisions
From Obsolete Lustre Wiki
Jump to navigationJump to search
No edit summary |
No edit summary |
||
| Line 1: | Line 1: | ||
The Metadata Performance Project is a joint project | The Metadata Performance Project is a joint project of the Lustre team, the ORNL LCE, and Cray initiated in early spring of 2009. Current areas of investigation include LNET, ''ptlrpc'', ''libcfs'', and the MDS server. The overall goal is to maximize single MDS performance through optimizations that utilize the multiple cores available in today's MDS machines. | ||
The project has the following primary goals: | The project has the following primary goals: | ||
*Understand current Lustre 1.6, 1.8, and 2.0 branch metadata performance characteristics. | *Understand current Lustre 1.6, 1.8, and 2.0 branch metadata performance characteristics. | ||
*Double the | *Double the metadata performance in the 2.x branch compared to the 1.6 baseline numbers. | ||
*Investigate SMP scaling improvements for the Lustre metadata stack. | |||
*Though introduction of clustered metadata for early exposure in Lustre 2.x and until the general release of CMD, ensure that Lustre metadata scales and performs as well as, or better than, GPFS metadata scaling. | *Though introduction of clustered metadata for early exposure in Lustre 2.x and until the general release of CMD, ensure that Lustre metadata scales and performs as well as, or better than, GPFS metadata scaling. | ||
Baseline metadata [http://wiki.lustre.org/images/1/13/MD-perf-comparison.pdf performance tests] were completed by the Lustre team on May 11, 2009. | Baseline metadata [http://wiki.lustre.org/images/1/13/MD-perf-comparison.pdf performance tests] were completed by the Lustre team on May 11, 2009. | ||
Revision as of 17:26, 15 December 2009
The Metadata Performance Project is a joint project of the Lustre team, the ORNL LCE, and Cray initiated in early spring of 2009. Current areas of investigation include LNET, ptlrpc, libcfs, and the MDS server. The overall goal is to maximize single MDS performance through optimizations that utilize the multiple cores available in today's MDS machines.
The project has the following primary goals:
- Understand current Lustre 1.6, 1.8, and 2.0 branch metadata performance characteristics.
- Double the metadata performance in the 2.x branch compared to the 1.6 baseline numbers.
- Investigate SMP scaling improvements for the Lustre metadata stack.
- Though introduction of clustered metadata for early exposure in Lustre 2.x and until the general release of CMD, ensure that Lustre metadata scales and performs as well as, or better than, GPFS metadata scaling.
Baseline metadata performance tests were completed by the Lustre team on May 11, 2009.
