Re: Large scale transparent proxying

From: Pete Yandell <>
Date: Mon, 18 Oct 1999 10:48:14 +1000

>> > On Thu, 14 Oct 1999, Pete Yandell wrote:
>> >
>> >> I'm running squid-2.2.STABLE4 under NetBSD 1.4.1 doing transparent
>> >> proxying without caching, and with about 200 concurrent users (I don't
>> >> have a figure for number of hits per second) the load gets really high
>> >> (85% CPU usage or thereabouts) on a Pentium II 450 with 256Mb of RAM.
>> >> A lot of that load seems to be system CPU time dealing with the
>> >> required NAT translation.
>> >>
>> >> It seems to me that load shouldn't really be high...what it's doing is
>> >> really pretty simple, especially given that it's not caching.
>> >
>> > Just a thought, but what sort of network card are you using?
>> Intel EtherExpress 10/100 (running at 100).
> That card shouldn't eat CPU time at all. I could suggest using LFS
> (instead of FFS) as the filesystem type for the cache dir. It's a lot
> better performing, but you'd do it at your own risk :). Maybe FFS with -o
> async,noatime.
> If your ipfilter rules are complex enough that'll contribute to the high
> load.

I'm not caching at all...I have some big parent proxies to take care
of that, so filesystem performance isn't the problem.

I have one ipfilter rule, so complexity is not the problem.

There are a few other weirdnesses happening too. I'm going to try a
few things and post something to the NetBSD lists as well. I'll post
back here when I know a little more.

Peter Yandell
Chief Technology Officer Pty Ltd
Received on Sun Oct 17 1999 - 18:56:15 MDT

This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 16:48:56 MST