Re: 2nd-biggest Squid performance problem?

Andres Kroonmaa wrote:

> > Now that that's going, what is Squid's next Big Performance Problem?
> cpu and latency pigs are:
> 1) diskio
> 2) networkio
> 3) ACLs
> in this order.
> Currently, in addition to networkio itself, poll overhead is damned.

diskio we know quite well how to address.. mostly a matter of
implementing. Adrian is running one development thread there, and I have
another one (at a very low priority).

as discussed there is ideas on how to start improving network I/O.
Mainly the eventio thread, and where it allows us to end up. Again, is
in serious need of development time.

> Disk performance is hardwired and you can't really overcome it. Disk
> overhead affects performance because cpu is consumed for nothing.
> But when lots of sessions are connected, 10K for eg., then things
> look different, dramatically. disk overhead seems totally nonissue.

What you can do is to make sure you don't spend disk I/O on nothing, as
the majority (or at least a very large portion) of the I/O for a cache
is today.

> Whichever provides less overhead under extreme conditions. Ideal
> targets are 1-5K reqs/sec at 10K sessions on generally affordable
> hardware. eventio is prerequisite to even think about it.
> eventio actually helps to think in direction of SMP too.

1K req/sec at 10K sessions should not be any major problems I think. 5K
req/s will for sure require some additional thought.

Regards
Henrik
Received on Fri Sep 28 2001 - 01:58:54 MDT