From: | Zeugswetter Andreas SB <ZeugswetterA(at)wien(dot)spardat(dot)at> |
---|---|
To: | "'The Hermit Hacker'" <scrappy(at)hub(dot)org>, Zeugswetter Andreas SB <ZeugswetterA(at)wien(dot)spardat(dot)at> |
Cc: | "'Mike Cianflone'" <mcianflone(at)littlefeet-inc(dot)com>, Hackers List <pgsql-hackers(at)postgresql(dot)org> |
Subject: | AW: AW: vacuum |
Date: | 2001-06-13 14:01:32 |
Message-ID: | 11C1E6749A55D411A9670001FA687963368324@sdexcsrv1.f000.d0188.sd.spardat.at |
Views: | Raw Message | Whole Thread | Download mbox | Resend email |
Thread: | |
Lists: | pgsql-hackers |
> The other question is, what are your startup parameters? What sort of
> shared memory buffer are you working with? I image that VACUUM does a
> flush to disk, so are 1000 items filling your buffer(s),
> causing an almost
> continue fsync to disk for each INSERT after that ... so the VACUUM is
> clearing the buffer(s) to disk, giving your space again to do quick
> INSERTs for 1000 records again?
Aah yes, you mean that the pages are only written during checkpoints ?
Then he would not need a vacuum, but a forced checkpoint.
That would actually impose, that we need yet another checkpoint initiating
event, e.g. do checkpoint when buffer is 75% dirty. Or better yet have
a daemon, that only writes out dirty pages from back to front of the LRU queue
until only a configurable amount of buffers is dirty.
Andreas
From | Date | Subject | |
---|---|---|---|
Next Message | Bruce Momjian | 2001-06-13 14:07:37 | Re: Indexing on a separate volume |
Previous Message | Bruce Momjian | 2001-06-13 14:01:00 | Re: Indexing on a separate volume |