Re: some profiling info

Arjan de Vet <Arjan.deVet@adv.IAEhv.nl> writes:

> Oskar Pearson:
>
> >Perhaps it's time we had a 'statd' (like unlinkd)? Note that 'oldselect'
> >is the same as 'select'. On that topic - why are there unlink calls at
> >all?
>
> I had a quick look at the source and the stat system call is mainly used
> during starting and stopping Squid AFAIK, especially when the swaplog file
> has to be rebuild. So these stat results look a little bit strange to me.
>
> I'm working on an updated and improved version of my Squid timing patch.
> What I see on a very busy cache (100,000 conns/hour) is that open(2) is
> *the* bottleneck (50-60% of the time :-() and we should have something
> like an opend or seperate processes which do all disk I/O so Squid doesn't
> stop during blocking disk operations.

Indeed. Spot on what I get on my 150,000 TCP/hour cache.

I've got an experimental patch for squid 1.NOVM.15 that splits some of
the opens into seperate threads. Been running a couple of days now,
and seems mostly stable, but ... :)

Built around the linuxthreads-0.6 package (i.e. clone() based).

Not 100% convinced that I shouldn't be fixing the 1.2beta stuff
instead, but I needed something working now...

Michael.
Received on Tue Jul 29 2003 - 13:15:44 MDT