Re: benchmarking squid on solaris/x86

On Wed, 20 Mar 2002, Andres Kroonmaa wrote:

> > I don't want to tweak TCP stuff too much. Ideally I'd like all
> > of the OSes to have similar TCP parameters as possible.
>
> I argue with such approach. Its not that ideal. TCP is integral
> part of kernel, and tuning kernel but rejecting tuning of TCP
> is not consistent.
> Using OS defaults makes obviously no sense. Picking arbitrary
> unified settings makes also little sense since different OSes
> implement different TCP options and get different hit while
> tweaking TCP params. In any case picked TCP params must have
> a good relation to reallife tunables. For very highperf cache
> you'd sometimes even have to resort to bending few standards.
>
> <snipped good stuff about sockets in TIME_WAIT state>
>
> I'd try to squeeze max performance out of each OS by whatever
> means, then analyse what has an impact and how acceptable is
> required tweaking.

I think the two approaches are orthogonal and equally valuable for
different types of sysadmins. The first approach -- stick with the
defaults if possible -- gives people an idea about Squid performance
on a given OS without spending time on [often risky and
protocol-violating] optimizations. The second approach -- optimize as
much as you can -- provides an upper bound of Squid performance on a
given OS.

So, if Duane has enough time and desire, he may investigate both
approaches. If not, he will probably pick one and upset about 50% of
sysadmins...

$0.02,

Alex.
Received on Wed Mar 20 2002 - 09:56:20 MST