Re: new vacuum is slower for small tables

"Pavel Stehule" <pavel(dot)stehule(at)gmail(dot)com> writes:

> Hello,
>
> I did small tests and I found so for small tables (less 1000 rows)
> VACUUM based on visibility maps are slower than old implementation
>
> it is about 5ms X 20ms

Hm, Presumably this is on a machine where the visibility map is entirely in
cache too. On a busy machine that's stealing a page from other table's cache.

The conservative thing to do would be to avoid creating a visibility map for
tables unless they're large enough that it's incontrovertibly enough of a
gain. Especially since speeding up short-lived vacuums doesn't seem terribly
interesting.

Without index-only scans it probably doesn't make much sense to create the
visibility map until we hit something like 32 pages or so. Vacuum should be
able to handle that many pages so fast that speeding it up seems pointless.

--
Gregory Stark
EnterpriseDB http://www.enterprisedb.com
Ask me about EnterpriseDB's 24x7 Postgres support!