Re: [grahame@ucs.uwa.edu.au: Patch to squid HEAD]

Second review after trying to understand the delay pool stuff..

I think we can drop the whole slowfd stuff from comm_select. It only seems to
be a very weak attempt in giving no_delay filedescriptors a higher priority
than delayed ones, but as such it is very weak as it only looks at a single
poll().

The change in delay_pools.c is fine, except maybe that it may be better to
use an array of int instead of char (less overhead).

Regards
Henrik

On Thursday 15 November 2001 14.50, Adrian Chadd wrote:
> Review?

> Adrian
>
> ----- Forwarded message from Grahame Bowland <grahame@ucs.uwa.edu.au> -----
>
> Subject: Patch to squid HEAD
> From: Grahame Bowland <grahame@ucs.uwa.edu.au>
> To: adrian@squid-cache.org
> X-Mailer: Evolution/0.99.0 (Preview Release)
> Date: 15 Nov 2001 12:40:43 +0800
>
>
> Hi Adrian,
>
> Attempting to get onto squid-dev, but until then I'll send patches to
> you :) This patch removes the dependency on the OS specified FD_SETSIZE
> when using polled I/O. Basically it removes fd_set from the delay pools
> code and replaces it with an array of chars, as that was the only thing
> tying us down to the FD_SETSIZE limit for Squid_MaxFD.
>
> We've been running with this on styx for a few days, since 1024
> filedescriptors was too small.
>
> Anyway, patch attached and I used diff -u this time :)
Received on Fri Nov 16 2001 - 09:59:26 MST