Re: Postgres performance Linux vs FreeBSD

On 2007-02-21, Mark Kirkwood <markir(at)paradise(dot)net(dot)nz> wrote:
> With respect to 'select count(*) from ...' being slower on FreeBSD,
> there are a number of things to try to make FreeBSD faster for this sort
> of query. Two I'm currently using are:
>
> - setting sysctl vfs.read_max to 16 or 32
> - rebuilding the relevant filesystem with 32K blocks and 4K frags

Be aware that increasing the filesystem block size above 16k is _known_
to tickle kernel bugs - there is a workaround involving increasing
BKVASIZE, but this isn't a config parameter and therefore you have to
patch the sources.

The symptom to look for is: largescale filesystem deadlocks with many
processes (especially syncer) blocked in "nbufkv" state.

--
Andrew, Supernews
http://www.supernews.com - individual and corporate NNTP services