On Tuesday February 10 2004 10:26, Tom Lane wrote:
>
> Giving PG half the RAM is counterproductive no matter what --- that
> pretty much guarantees that every page that's in RAM will be in RAM
> twice, once in PG buffers and once in kernel buffers. The two
> reasonable schools of thought are (1) to make PG buffers relatively
> small and let the kernel do the bulk of the buffering, or (2) to give
> PG most of the RAM and expect it to do the bulk of the buffering.
>
> Past experience has been that theory (1) generally wins. Jan's recent
> work on ARC buffering may make things more interesting, though.
So, pursuing theory (1) in 'the time before ARC', assuming you have a
dedicated box with little or no non-db competition for resources, why give
PG anymore than is absolutely required (max_connections*2 or 16)? Why not
just let the kernel reign as completely as possible? Is that what you mean
by "relatively small"?
TIA.