Re: 8.4 open item: copy performance regression?

On Fri, Jun 19, 2009 at 07:49:31PM +0200, Stefan Kaltenbrunner wrote:
> Tom Lane wrote:
>> Just eyeing the code ... another thing we changed since 8.3 is to enable
>> posix_fadvise() calls for WAL. Any of the complaints want to try diking
>> out this bit of code (near line 2580 in
>> src/backend/access/transam/xlog.c)?
>> #if defined(USE_POSIX_FADVISE) && defined(POSIX_FADV_DONTNEED)
>> if (!XLogArchivingActive() &&
>> (get_sync_bit(sync_method) & PG_O_DIRECT) == 0)
>> (void) posix_fadvise(openLogFile, 0, 0, POSIX_FADV_DONTNEED);
>> #endif
>
> ok after a bit of bisecting I'm happy to announce the winner of the
> contest:
>
> http://archives.postgresql.org/pgsql-committers/2008-11/msg00054.php
>
> this patch causes a 25-30% performance regression for WAL logged copy,
> however in the WAL bypass case (maybe that was what got tested?) it results
> in a 20% performance increase.
>
> the raw numbers using the upthread posted minimal postgresql.conf are:
>
> post patch/wal logged: 4min10s/4min19/4min12
> post patch/wal bypass: 1m55s/1m58s/2m00
> prepatch/wal logged: 2m55s/3min00/2m59
> prepatch/wal bypass: 2m22s/2m18s/2m20s
>
>
> Stefan
>

Great! Maybe just increasing the size of the BULKWRITE ring,
possibly as a function of the shared_memory is all that is
needed. 256kB is the currently coded ring_size in storage/buffer/freelist.c

Ken