[LU-2254] Lustre metadata performance tracker

• Status: Open
• Project: Lustre
• Type: Task
• Priority: Minor
• Reporter: Liang Zhen (Inactive)
• Assignee: WC Triage
• Resolution: Unresolved

Description

Create a new ticket to track history data and technical discussions of Lustre metadata performance.

Comments

[Liang Zhen (Inactive)]

Here is 2.3 metadata performance demonstration for openSFS:
http://jira.whamcloud.com/secure/attachment/11805/opensfs_smp_demonstration.pdf

Here is patched mdtest: (http://jira.whamcloud.com/secure/attachment/11995/mdtest_lustre_1.8.3.tgz)

• user can specify stripecount of file
• can run for mknod
• user can use range expression to specify multiple paths

[Liang Zhen (Inactive)]

still use the same environment, change online CPU number of MDS and rerun some tests
(http://jira.whamcloud.com/secure/attachment/12013/smp_perf_fat_intel_2.pdf)

[Liang Zhen (Inactive)]

AMD server as MDS (a little slower than the intel server) which has 16 cores, rerun some tests with both 2.2 and 2.3
Also, because AMD server doesn't have SSD, so use 4G ramdisk ask journal device.
(http://jira.whamcloud.com/secure/attachment/12014/smp_perf_fat_amd.pdf)

[Liang Zhen (Inactive)]

again, the same environment, intel server. rerun tests with different external & internal journal. (2.3 only)
(http://jira.whamcloud.com/secure/attachment/12015/smp_journal_perf_fat_intel_2.pdf)

[Liang Zhen (Inactive)]

raw data of previous three tests
http://jira.whamcloud.com/secure/attachment/12017/cpu_and_journal_log.tgz

[Shuichi Ihara (Inactive)]

Hi Liang,
This is quite interesting. What do you mean multi-mount lustre on the client?