On Sat, 24 Apr 2004 15:48:19 -0400, "Shea,Dan [CIS]" <Dan.Shea@ec.gc.ca>
wrote:
>Manfred is indicating the reason it is taking so long is due to the number
>of dead tuples in my index and the vacuum_mem setting.
<nitpicking>
Not dead tuples in the index, but dead tuples in the table.
</nitpicking>
>The last delete that I did before starting a vacuum had 219,177,133
>deletions.
Ok, with vacuum_mem = 196608 the bulk delete batch size is ca. 33.5 M
tuple ids. 219 M dead tuples will cause 7 index scans. The time for an
index scan is more or less constant, 60000 seconds in your case. So
yes, a larger vacuum_mem will help, but only if you really have as much
*free* memory. Forcing the machine into swapping would make things
worse.
BTW, VACUUM frees millions of index pages, is your FSM large enough?
Servus
Manfred
