| From: | Jim Nasby <decibel(at)decibel(dot)org> |
|---|---|
| To: | Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us> |
| Cc: | Josh Berkus <josh(at)agliodbs(dot)com>, pgsql-hackers(at)postgresql(dot)org |
| Subject: | Re: Bug: Buffer cache is not scan resistant |
| Date: | 2007-03-07 00:38:44 |
| Message-ID: | AF45B115-9BF5-47F6-8C05-B2459EE84D4F@decibel.org |
| Views: | Raw Message | Whole Thread | Download mbox | Resend email |
| Thread: | |
| Lists: | pgsql-hackers |
On Mar 6, 2007, at 12:17 AM, Tom Lane wrote:
> Jim Nasby <decibel(at)decibel(dot)org> writes:
>> An idea I've been thinking about would be to have the bgwriter or
>> some other background process actually try and keep the free list
>> populated,
>
> The bgwriter already tries to keep pages "just in front" of the clock
> sweep pointer clean.
True, but that still means that each backend has to run the clock-
sweep. AFAICT that's something that backends will serialize on (due
to BufFreelistLock), so it would be best to make StrategyGetBuffer as
fast as possible. It certainly seems like grabbing a buffer off the
free list is going to be a lot faster than running the clock sweep.
That's why I think it'd be better to have the bgwriter run the clock
sweep and put enough buffers on the free list to try and keep up with
demand.
--
Jim Nasby jim(at)nasby(dot)net
EnterpriseDB http://enterprisedb.com 512.569.9461 (cell)
| From | Date | Subject | |
|---|---|---|---|
| Next Message | Jim Nasby | 2007-03-07 00:43:43 | Re: Bug: Buffer cache is not scan resistant |
| Previous Message | Gavin Sherry | 2007-03-07 00:29:47 | Re: PostgreSQL - 'SKYLINE OF' clause added! |