Mailing List Archive

What could be the problem
I switch the server at Cardiff www.cs.cf.ac.uk to 0.8.14 the other
day and put in -DMINIMAL_DNS 'cos we've given up on recording that
crap on the fly.

Today the server is very unhappy. Connection times are very long
even from local machines. There's a machine right next to it running
0.8.7 with the DNS on but with much less traffic. Responses there are
fast and reliable.

I switched back to 0.8.12 with DNS on and the same delays occur.. it takes
just a few seconds after startup or SIGHUP, then the delays seem to stick.
A reboot failed to fix the problem.


The load on the machine is lower than normal due to the reduction in
incoming requests. Child prcesses seem to come and go without problems.

Anyone seen this kind of problem before?

I just noticed that in the logfile that there are occasional long pauses
between requests, 20-30 seconds at times, when there should be about 2
conntections per second.

When I connect from a local machine with Lynx, it ocassionally says
"looking up host www.cs.cf.ac.uk" even though it knows all about the
machine which is only a few feet away.

Has Sprint been tinkering with the net today?


rob
Re: What could be the problem [ In reply to ]
> Has Sprint been tinkering with the net today?

I'm sure now that it's the listen queue that's filling up due to
crappy networks. AFAIK, the sysadmin at Cardiff failed to find the
necessary SunOs config setting to up the number from 5 to a realistic
100.

Most of the traffic we're getting comes from the US, and for some
unkown reason, today they all seem to want to look at Pamela Anderson's
info in the movie database.

:-(