On Wed, 27 Jun 2001, Philip Molter wrote:
> On Wed, Jun 27, 2001 at 06:58:18PM -0400, Bruce Momjian wrote:
> : My guess on this one is that Solaris is slower for PostgreSQL because
> : process switching is _much_ heavier on Solaris than other OS's. This is
> : because of the way they implemented processes in SVr4. They got quite
> : heavy, almost requiring kernel threads so you weren't switching
> : processes all the time.
> :
> : In a sense threads were a solution to a process bloating problem.
> : Linux/BSD have much lighter processes and hence work better for
> : PostgreSQL. Again, this is only a guess.
> :
> : MySQL does more stuff with threads while PostgreSQL switches process
> : because each backend is a process.
>
> Does more stuff with threads? It does all stuff with threads. Your
> guess was our guess, which is why we tried shoving the thing over to a
> Linux box. Now if I only I could figure out why kernel CPU usage keeps
> going up incrementally over time (went from roughly a 5% average to a
> 16% average in two days) the more we run the system. All signs are
> pointing to postgres.
Or Linux. Try 2.4 kernel which is far better as far as SMP goes.
Regarding threads, there's a very good quote from Linus:
"Solution to slow process switching is fast process switching, not another
kernel abstraction".
> VACUUM ANALYZE-ing the tables used to reduce it back down, but now, it
> doesn't appear to be as effective (might go from 16% back down to
> 13%). Anyone know what causes that, and better yet, anyone know how to
> fix it? We see similar behavior under Solaris.
Look at linux profiling tools.
-alex
