[SQU] Run away memory problem

We have three Squid servers configured the same that are running
squid-2.3.STABLE4 on RH Linux 2.2.12-20smp, 256MB of RAM and 4G for squid
cache.

We have our cache_mem set to 16 MB, and memory_pools turned off. We have
used both Linux mallac and dl-malloc with the same results. We are
currently using replacement_policy LFUDA.

The problem is that every day one of our three servers starts to use up all
of the memory and go into swap space. When looking at the Memory
Utilization section in the Cache Manager the "Store Mem Buffer" is taking up
all the memory around 200MB. (Store Mem Buffer 4096 49456 197824 197824)
Shouldn't the cache_mem parameter limit this section of memory to 16MB?

Anther oddity is that when this happens the Byte Hit Ratios goes negative on
the General Runtime Information page:
        Byte Hit Ratios: 5min: -37.1%, 60min: -32.3%

Also, when we shutdown squid (via squid -k shutdown) to clear up this
problem squid will not stop. We have manually shut squid down via kill.

Has this happened to anyone else? and is there anything else I can do to
figure out what is going on?

Thanks,
-Dan

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