Re: Bug: Buffer cache is not scan resistant - Mailing list pgsql-hackers

From Luke Lonergan
Subject Re: Bug: Buffer cache is not scan resistant
Date
Msg-id C3E62232E3BCF24CBA20D72BFDCB6BF802CFC0D7@MI8NYCMAIL08.Mi8.com
Whole thread Raw
In response to Re: Bug: Buffer cache is not scan resistant  (Mark Kirkwood <markir@paradise.net.nz>)
List pgsql-hackers
Hi Mark,

> lineitem has 1535724 pages (11997 MB)
>
> Shared Buffers  Elapsed  IO rate (from vmstat)
> --------------  -------  ---------------------
> 400MB           101 s    122 MB/s
>
> 2MB             100 s
> 1MB              97 s
> 768KB            93 s
> 512KB            86 s
> 256KB            77 s
> 128KB            74 s    166 MB/s
>
> I've added the observed IO rate for the two extreme cases
> (the rest can be pretty much deduced via interpolation).
>
> Note that the system will do about 220 MB/s with the now
> (in)famous dd test, so we have a bit of headroom (not too bad
> for a PIII).

What's really interesting: try this with a table that fits into I/O
cache (say half your system memory), and run VACUUM on the table.  That
way the effect will stand out more dramatically.

- Luke



pgsql-hackers by date:

Previous
From: Mark Kirkwood
Date:
Subject: Re: Bug: Buffer cache is not scan resistant
Next
From: Gregory Stark
Date:
Subject: Re: Bug: Buffer cache is not scan resistant