Page History
Lustre 2.2 is scheduled for release was released at the end of Q1 2012. This is its anticipated support matrix.
Server Support | Client Support |
|---|---|
RHEL 6, | RHEL 6, |
As well as ongoing stability improvements, there are a number of new feautures features in the Lustre 2.2 release.
Feature | Summary | JIRA | Notes | ||||
|---|---|---|---|---|---|---|---|
Statahead | Performance improvement for ls -l. This feature accelerates sequential directory traversals by pre-fetching object attributes from the MDS. Fan Yong presented this at LUG 2011. | ||||||
Asynchronous glimpse lock (AGL) | AGL is an optimization of Statahead. It asynchronously fetches size and block attributes from the OSTs for better ls -l performance. Fan Yong from Whamcloud presented this feature at LUG 2011. | ||||||
Client parallel checksums | Improves speed of client checksums by distributing the workload over all cores in a CPU by using the ptlrpcd feature. Also improves the accuracy by making checksums work properly for memory mapped pages. | ||||||
Imperative Recovery | IR increases speed of recovery for larger deployments by actively notifying clients of server restarts for faster reconnections than the previous timeout mechanism. This feature was presented at LUG 2011 by Jinshan Xiong. | ||||||
Large xattrs | Increases AKA Wide Striping. This feature increases the maximum stripe count from 160 to 2000 and was presented at LUG 2011 by Oleg Drokin. This feature effectively increases the maximum file size under ldiskfs to 2000 * 16TB = 32PB. | ||||||
Mds-survey | Tool This tool adds Metadata support in echo_client for benchmarking metadata performance. Now the MDS can be performance tested without requiring any clients to create metadata operations. This feature also includes a user friendly script similar to obdfilter_survey for easier testing. |
| |||||
Multi threaded ptlrpcd | Performance enhancement | N\A |
| This performance enhancement implements a general ptlrpcd pool. The ptlrpcd threads in this pool are shared by all the asynchronous RPCs on the client, such as BRW requests, data checksums, async glimpse locks, statahead, etc. | |||
OSD API | The OSD restructuring project will allow alternate backend file systems to be used by Lustre. This feature is the first step in moving the new code to a released version of Lustre, and adds the OSD API to the OSS stack. This was presented to the EOFS LUG 2011 by Alex Zhuravlev. | OSD API |
|
| |||
Parallel directory operations (pDirOps) | Performance improvement for operations performed on the same directory | pDirOps introduces more granular locking on ext4 directory structures, thus allowing more concurrent operations simultaneously. This feature was presented at LUG 2011 by Liang Zhen. | |||||
Quota protocol/code improvements |
| This feature prepares the Lustre quota protocol for future quota changes which allow clients to get quota directly from OSTs instead of the MDS. It also does some reorganization to remove old/dead code, and adds new tests for quota accounting. |
| ||||
Layout Lock Part 1 | This patch creates a generation number for the file layout. This number will be used in conjunction the layout lock to allow clients to detect when the file layout has changed. This is an important feature for HSM and is the first of many HSM features to be landed in subsequent releases. |
Overview
Content Tools