Re: reindex/vacuum locking/performance?

On Sun, Oct 05, 2003 at 07:32:47PM -0400, Neil Conway wrote:

> been pointed out numerous times on -hackers and in the literature, using
> LRU for a DBMS shared buffer cache is far from optimal, and better
> algorithms have been proposed (e.g. LRU-K, ARC). We could even have the
> VACUUM command inform the bufmgr that the pages it is in the process of
> reading in are part of a seqscan, and so are unlikely to be needed in
> the immediate future.

Hey, when that happens, you'll find me first in line to praise the
implementor; but until then, it's important that people not get the
idea that vacuum is free.

It is _way_ imporved, and on moderately loaded boxes, it'salmost
unnoticable.  But under heavy load, you need to be _real_ careful
about calling vacuum.  I think one of the biggest needs in the AVD is
some sort of intelligence about current load on the postmaster, but I
haven't the foggiest idea how to give it such intelligence.

A

--
----
Andrew Sullivan                         204-4141 Yonge Street
Afilias Canada                        Toronto, Ontario Canada
<andrew@libertyrms.info>                              M2P 2A8
                                         +1 416 646 3304 x110