Re: I/O increase after upgrading to 8.3.5

From: "Kevin Grittner" <Kevin(dot)Grittner(at)wicourts(dot)gov>
To: "Alexander Staubo" <alex(at)bengler(dot)no>
Cc: "pgsql-performance" <pgsql-performance(at)postgresql(dot)org>
Subject: Re: I/O increase after upgrading to 8.3.5
Date: 2009-02-13 17:35:29
Message-ID: 49955B00.EE98.0025.0@wicourts.gov
Views: Raw Message | Whole Thread | Download mbox | Resend email
Thread:
Lists: pgsql-performance

>>> Alexander Staubo <alex(at)bengler(dot)no> wrote:
> Kevin Grittner <Kevin(dot)Grittner(at)wicourts(dot)gov> wrote:
>> Could you show the non-commented lines from old and new
>> postgresql.conf files, please?
>
> Attached. The differences are not performance-related, as far as I
> can see, aside from the additional of "synchronous_commit = off".

You should definitely set effective_cache_size.

If you still see the problem after that, I suggest testing different
settings for:

bgwriter_lru_maxpages
bgwriter_lru_multiplier
checkpoint_segments
checkpoint_timeout
checkpoint_completion_target

Both the checkpoint process and the background writer changed quite a
bit, and these are the main knobs for tuning the new version.

It's possible that under 8.2 your hint bit writes were being combined
with other writes due to caching, and differences in the timings now
have the hint bit writes happening too long after the initial write to
get that benefit. If that's the case, the counterintuitive step of
making PostgreSQL more aggressive about writing to the OS cache might
reduce your disk write I/O. Making it less aggressive might allow
them to combine in the PostgreSQL buffers before making to the OS
cache.

-Kevin

In response to

Responses

Browse pgsql-performance by date

  From Date Subject
Next Message Chris Browne 2009-02-13 17:37:21 Re: scheduling autovacuum at lean hours only.
Previous Message Tom Lane 2009-02-13 17:20:58 Re: [PERFORM] GIST versus GIN indexes for intarrays