Thread: Advice on testing buffer spin lock contention

Hi all,

We are working on a patch which targets the overhead of spinlock in buffer cache.In this regard, we need to exercise
bufferspinlock contention on a 16 core machine. 

Could anyone please suggest some good methods for exercising the same?

Regards,

Atri

Sent from my iPad

> You will want a select only workload in which all data fits in
> shared_buffers, and that doesn't do a round trip to some external driver
> program for each row selected.
>
> I proposed a patch for pgbench to add a new transaction of this nature last
> year under the subject "pgbench--new transaction type".  Heikki also
> described how you can make a custom pgbench -f file to accomplish the same
> thing without having to change and recompile pgbench.
>
>
> There was also a length discussion about nailing certain pages so they don't
> need to get pinned and unpinned all the time, under "9.2beta1, parallel
> queries, ReleasePredicateLocks, CheckForSerializableConflictIn in the
> oprofile"
>

Thanks a ton, we will try it out.

Regards,

Atri


--
Regards,

Atri
l'apprenant