> Kari Lavikka <tuner@bdb.fi> writes:
> > However, those configuration changes didn't have significant effect
to
> > oprofile results. AtEOXact_CatCache consumes even more cycles.
>
> I believe I've fixed that for 8.1.
Relative to 8.0, I am seeing a dramatic, almost miraculous reduction in
CPU load times in 8.1devel. This is for ISAM style access patterns over
the parse/bind interface. (IOW one record at a time, 90% read, 10%
write).
Relative to commercial dedicated ISAM storage engines, pg holds up very
well except in cpu load, but 8.1 is a huge step towards addressing that.
So far, except for one minor (and completely understandable) issue with
bitmap issues, 8.1 has been a stellar performer. Also great is the
expansion of pg_locks view (which I didn't see mentioned in Bruce's TODO
list, just FYI).
Merlin