Dimitri <dimitrik.fr@gmail.com> wrote:
> PostgreSQL: 8.3.7 & 8.4
> Server: Sun M5000 32cores
> OS: Solaris 10
Does that have a battery backed RAID controller? If so, is it
configured for write-back? These both help a lot with smoothing
checkpoint I/O gluts.
We've minimized problems by making the background writer more
aggressive. 8.3 and later does a better job in general, but we've
still had to go with:
bgwriter_lru_maxpages = 1000
bgwriter_lru_multiplier = 4.0
> shared_buffers = 12000MB
You might want to test with that set to something much lower, to see
what the checkpoint delays look like. We've found it best to use a
small (256MB) setting, and leave caching to the OS; in our
environment, it seems to do a better job of scheduling the disk I/O.
YMMV, of course.
-Kevin