Greg Smith <gsmith@gregsmith.com> wrote:
> After a few months of staring at this data, I've found averages like that
> misleading. The real problem areas correlate with the peak pages written
> at any one checkpoint. Lowering that value is really the end-game for
> optimizing the background writer, and the peaks are what will nail you
> with a nasty fsync pause at checkpoint time.
If you've already had some technical knowledge to lead best settings from
activity logs, could you write it down in the codes? I hope some kinds of
automatic control features in bgwriter if its best configurations vary by
usages or activities.
BTW, I'm planning two changes in bgwriter.
[Load distributed checkpoint] http://archives.postgresql.org/pgsql-patches/2007-02/msg00522.php [Automatic
adjustmentof bgwriter_lru_maxpages] http://archives.postgresql.org/pgsql-patches/2007-03/msg00092.php
I have some results that if we have plenty of time for checkpoints,
bgwriter_all_maxpages is not a so important parameter because it is
adjusted to "shared_buffers / duration of checkpoint".
Also, my recommended bgwriter_lru_maxpages is "average number of
recycled buffers per cycle", that is hardly able to tune manually.
Regards,
---
ITAGAKI Takahiro
NTT Open Source Software Center
