[LU-9905] cpu_npartitions is not configured properly Created: 23/Aug/17  Updated: 22/Nov/17  Resolved: 10/Oct/17

Status: Resolved
Project: Lustre
Component/s: None
Affects Version/s: Lustre 2.10.0
Fix Version/s: None

Type: Bug Priority: Minor
Reporter: Shuichi Ihara (Inactive) Assignee: Dmitry Eremin (Inactive)
Resolution: Not a Bug Votes: 0
Labels: None

Severity: 3
Rank (Obsolete): 9223372036854775807

 Description   

cpu_npartitions=10 is configured manually, but after re-loaded module, never cpu_partition_table changed.

[root@oss01 ~]# umount -t lustre -a; lustre_rmmod

[root@oss01 ~]# cat /etc/modprobe.d/lustre.conf 
options lnet networks="o2ib10(ib0)"
options libcfs cpu_npartitions=10

[root@oss01 ~]# mount -t lustre 10.0.10.167@o2ib10:10.0.10.168@o2ib10:/scratch /scratch -o localflock 
[root@oss01 ~]# cat /proc/sys/lnet/cpu_partition_table 
0	: 0 2 4 6 8 10 12 14 16 18
1	: 1 3 5 7 9 11 13 15 17 19


 Comments   
Comment by Peter Jones [ 23/Aug/17 ]

Dmitry

Can you please advise on this one?

Thanks

Peter

Comment by Dmitry Eremin (Inactive) [ 21/Sep/17 ]

Sorry for delay. This behavior is expected. Resently we changed the default value for CPU partitions.

https://review.whamcloud.com/22507

So, now cpu_pattern is specified and have default value "N". But if cpu_pattern is specified, the cpu_npartitions will be ignored.

You can return to previous behavior by spefifing empty value for cpu_pattern.

So, please use the following configuration:

options libcfs cpu_pattern="" cpu_npartitions=10

Comment by Shuichi Ihara (Inactive) [ 10/Oct/17 ]

ok, that workaround works. you can close this ticket.
However, after patch 22507 landed, default partition configuation for MDS caused huge performance regression. please check LU-9998.

Generated at Sat Feb 10 02:30:21 UTC 2024 using Jira 9.4.14#940014-sha1:734e6822bbf0d45eff9af51f82432957f73aa32c.