[SQU] Incredibly slow httpd accel

Resources are not an issue for this -- the box doesn't swap at all, trivial CPU
usage, 100baseT network, 10% FD usage.

We have 4 squid httpd accelerators on a busy network (4.4 mil. hits/day). I
used to have all 4 handling all requests but decided to pull more traffic away
from apache by leaving three as the front line and using the 4th as a parent
cache for binaries (zip, exe, hqx, etc).

In principal, it works fine. The problem is I've noticed cachable, but not
currently cached, requests often move incredibly slowly. They usually start out
well, but slow down to 5k/s (on 100baseT) and basically stop. It's
not communication between the squid boxes, either. I can telnet a request
straight to the parent from the parent (telnet localhost 80 ...) and it will do
the same thing.

This does not affect all requests, though. It only seems to affect requests
which have already been requested by others. Based on my observations this
evening, it appears the first client to request a file sets the pace for any
subsequent requests, even if the later ones are orders of magnitude faster.

If that's the case (and if it isn't, I'd like to find out what it is), what can
I do to fix it?

Thank you in advance
        -- Brian

--
To unsubscribe, see http://www.squid-cache.org/mailing-lists.html