On Tue, 23 Mar 2004, Tom Lane wrote:
> Bill Moran <wmoran(at)potentialtech(dot)com> writes:
> > Was this true for some previous version? I could have swore I read somewhere
> > that vacuum_mem had to be set high enough or vacuum wouldn't be able to clean
> > everything up (aside from anything locked in transactions).
>
> Nope, never been the case.
>
> > Is performance the only reason for increasing vacuum_mem?
>
> Yes.
Maybe Bill's thinking of the fsm settings and regular vacuums and the
limitations on how many tuples can be reclaimed by regular vacuuming being
tied to fsm settings?