On Tue, 2005-08-23 at 11:43 +0800, Adrian Chadd wrote:
> Hiya everyone,
>
> Over the last two years I've received a few emails, here and there,
> hinting at optimisations various people have done to squid-2.5 to make
> it run "faster". I'd like to start implementing some of these, if
> they're not too intrusive, into squid-2.5 just to try and keep it
> reasonably useful whilst others hammer on -3 and also so we have
> a list of places to look and things to keep in mind when -3 is stable.
>
> You don't have to post them publicly, I'm quite happy to summarise
> what I receive here.
>
> One thing I remember, but I can't find in my current mailbox,
> was an email someone sent a while ago outlining that reducing the
> number of s*printf()s in the store IO path helped reduce CPU usage.
>
> What other suggestions do y'all have?
OS interface. My production systems used to spend 50% of the time within
the kernel - after all the possible sysctl tweaks had been applied.
kqueue didn't seem to help - possibly due to inadequate system
configuration, I haven't tested epoll (OS was linux).
I won't comment on making the changes being useful or correct given the
freeze policy on 2.5; but if you do please make a note of the changes
somewhere, the developer site or on the wiki or wherever ;)
Kinkie
