Joe Conway wrote:
> Tom Lane wrote:
>
>> This is completely untrue. Increasing vacuum_mem will likely make
>> things faster on large tables (by avoiding the need for multiple passes
>> over the indexes). It will not change the end result though.
>
>
> I can attest to that, based on very recent empirical evidence. On a 28
> million row table, I saw something like 33% speed-up in going from 256MB
> to 320MB for vacuum_mem.
>
> Joe
When I finally ran the vacuum full I set vacuum_mem to 1 gig, but it
never used more than 20 meg. Took 40 minutes btw