[LU-11804] Auto tune cpu parttions for high IOPS workload Created: 18/Dec/18  Updated: 18/Dec/18

Status: Open
Project: Lustre
Component/s: None
Affects Version/s: None
Fix Version/s: None

Type: Improvement Priority: Minor
Reporter: Shuichi Ihara Assignee: WC Triage
Resolution: Unresolved Votes: 0
Labels: None

Attachments: PNG File SingleClient-4K-RandomIOPS.png    
Rank (Obsolete): 9223372036854775807

 Description   

There are number of IOPS related performance impromvenss in Lustre-2.12 that are not only server side, but also client side. Attached resutls is 4K random read IOPS on single client.
There is 40% performance improvmtns with default client setting in lustre-2.12, but after tune incresing of cpu partitions, IOPS bumps up a lot.
This is great improvements, but I am still wondering if we should have auto tuning of cpu partition on such high IOPS workload with out of the box.

 Comments   
Comment by Andreas Dilger [ 18/Dec/18 ]

Is the increased performance only due to increasing the number of ptlrpcd threads, to handle more concurrent RPC sending/checksums?

Generated at Sat Feb 10 02:47:04 UTC 2024 using Jira 9.4.14#940014-sha1:734e6822bbf0d45eff9af51f82432957f73aa32c.