SWAPFAIL again

Hi,

We are having problems with squid 2.3 stable 3. One indication of the problem is excessive
amount of TCP_SWAPFAIL_MISS entries in the access.log. The other, more frustrating, is
substantial slowing down of the cache.

The setup is as follows:

Platform: PII with FreeBSD 3.2-RELEASE. Squid was compiled with:

--enable-icmp --enable-delay-pools --enable-snmp --enable-cache-d
igests --enable-err-language=Russian-koi8-r --enable-heap-replacement

I have also appliend the patch: squid-2.3.stable3-storeExpiredReferenceAge.patch
I am using two swap dirs of type ufs. Squid is configured to use LFUDA replacement
algorithm.

The reported problem disappears when the cache dirs are removed and reinitialized. But then
the situation repeats after about a day or so (depends on average load). The slowdown is so
substantial, that mrtg, monitoring squid activities times out on SMNP requests. The CPU
load goes from normal 18% down to less then 1% when this happens. At the same time, the
network traffic, generated by squid goes down very significantly too. fsck does not reveal
any problems with those filesystems and the OS doesn't complain either.

This cache server is the primary cache for large Regional network and all problems with the
cache are very painful to the users.

Does anyone have any ideas how to cure this situation?

TIA,

Igor Alekseev
Yaroslavl Regional Network cache admin.
Received on Wed Jun 07 2000 - 00:18:10 MDT