[OT] linux 3.10 kernel will improve ipc,sysv semaphore scalability

[Andrea Suisani]

Hi all,

Just to let the community know that probably Linux kernel 3.10
will include some work to make sysv semaphore code more scalable.
Since Postgres use this kernel features I've thought to post it
here just for the sake of sharing a good news:

The thread where Rik van Riel posts the v3 patch series:

http://thread.gmane.org/gmane.linux.kernel/1460761

Here http://thread.gmane.org/gmane.linux.kernel/1460761/focus=1461794
Davidlohr Bueso (patches co-author) post an interesting results
comparing multiple runs of "[his] Oracle Swingbench DSS" workload.
Reported here for brevity:

TPS:
100 users: 1257.21 (vanilla) 2805.06 (v3 patchset)
400 users: 1437.57 (vanilla) 2664.67 (v3 patchset)
800 users: 1236.89 (vanilla) 2750.73 (v3 patchset)

ipc lock contention:
100 users: 8,74% (vanilla) 3.17% (v3 patchset)
400 users: 21,86% (vanilla) 5.23% (v3 patchset)
800 users 84,35% (vanilla) 7.39% (v3 patchset)

as soon as I've a spare server to play with and a little
of time I will give it a spin.

Andrea

[Shaun Thomas]

Post by Andrea Suisani
100 users: 1257.21 (vanilla) 2805.06 (v3 patchset)
400 users: 1437.57 (vanilla) 2664.67 (v3 patchset)
800 users: 1236.89 (vanilla) 2750.73 (v3 patchset)

Wow, I like the look of that. I wonder how this reacts with disabled
autogrouping and increasing sched_migration_cost. If the completely fair
scheduler has less locking contention with this patch-set, those tweaks
may not even be necessary. I need to see if I can find a system to test on.
--
Shaun Thomas
OptionsHouse | 141 W. Jackson Blvd. | Suite 500 | Chicago IL, 60604
312-676-8870
***@optionshouse.com

______________________________________________

See http://www.peak6.com/email_disclaimer/ for terms and conditions related to this email

[Andrea Suisani]

Hi

Post by Shaun Thomas

Post by Andrea Suisani
100 users: 1257.21 (vanilla) 2805.06 (v3 patchset)
400 users: 1437.57 (vanilla) 2664.67 (v3 patchset)
800 users: 1236.89 (vanilla) 2750.73 (v3 patchset)

Wow, I like the look of that.

Me too.

Post by Shaun Thomas
I wonder how this reacts with disabled autogrouping and increasing
sched_migration_cost. If the completely fair scheduler has less locking
contention with this patch-set, those tweaks may not even be necessary.
I need to see if I can find a system to test on.

Yep, I'm eager to test it too, but unfortunately I have
no spare server to use. Probably early next I'll test it
on my workstation.

Andrea