Re: swap log

From: Henrik Nordstrom <hno@dont-contact.us>
Date: Thu, 29 Nov 2001 09:43:07 +0100

Adrian Chadd wrote:

> Now, in actual fact, the current swap logs are written to disk
> via the write() method (ie not via an async IO operation) which
> may block anyway. I don't know how this contributes to squid's
> overall performance, but I don't think that call blocks much.

It may, but most often not.

There are two conditions a (small) append write on a active file may
block under:

a) Memory pressure, requiring the write to wait for a new free page when
crossing a page boundary

b) Pagein of filesystem metadata required when allocating a new disk
block.

The first is not very likely to happen unless you are seriously pushing
the disks too hard, or trying to write a lot of data (which is not the
case for swap.state).

Likewise for the second. The pages is almost always paged in already.

Regards
Henrik
Received on Thu Nov 29 2001 - 01:48:48 MST

This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 16:14:39 MST