RE: Squid on Solaris and Intel

At 10:40 AM 10/28/98 -0700, Jeff Madison wrote:
>It does appear to be a memory leak but the memory does not free up even
>after I stop squid. The only way to reclaim the memory is to restart the
>system. I've tried both the default malloc and dlmalloc. The default
>malloc is much faster but the memory disappears in a fraction of the time.
>
>Jeff Madison
>Systems Engineer
>(801)924-0900 x 101

I'm tempted to say that Solaris runs into trouble here. Stopping a process
should free all resources allocated to that process. Apparently this doesn't
work in your case. Disk buffers??? Has anyone else noticed this behaviour?

I have 2 proxies under my control running Solaris: 1x Sol2.5 on an Ultra1 and
1x Sol2.6 on an Enterprise server (3000 I think, I haven't seen the box
because
I do full remote control and it's 400km from here...) and both don't show this
problem (I do use dlmalloc). It might be that you are triggering some bug with
your specific config. You might want to ask Sun about this but you could also
decide to go to Linux. I hear good stories about squids on Linux.

Marc

At 10:57 AM 10/27/98 -0700, Jeff Madison wrote:
[...]
>ith -enable-poll --enable-async-io --enable-dlmalloc -enable-ipf-transpar
e
>nt. I am using the default squid.conf with some very minor modifications
>such as changing the http port to 8080. I am putting the system under a
>some what heavy load but not any thing it should not be able to handle.
The
>problem occurs after the system has been running for about an hour. The
>system just runs out of memory and becomes extremely slow. The squid
>process its self is only using 20-30 MB or the memory. I would appreciate
>any suggestions for optimization or some type of solution to this problem.

---------------------------------------------------------------------
Marc van Selm
NATO C3 Agency
Communication Systems Division, A-Branch
Tel: +31 70 3142454
---------------------------------------------------------------------
Private: selm@cistron.nl, selm@het.net, http://www.cistron.nl/~selm

     "Push to test." <click> "Release to detonate."
Received on Thu Oct 29 1998 - 00:54:47 MST