Re: DU 4.0 and squid 1.1.10?

On Wednesday 16 July 97, at 19 h 56, the keyboard of
jprovo@ma.ultranet.com wrote:

> We're using DU 4.0, and had been working with squid 1.1.10, trying to
> finally get off of harvest. Unfortunately, even when built with real
> (gnu) malloc libs, squid seems to bloat its footprint beyond the
> capacities of the machine on which it is running.

Could you provide more details? Output of cache manager's "info", 'ps',
'vmstat', hardware configuration?

> - anyone with DU4.0 on the list and active?

The Renater national cache <http://cache.cnrs.fr/> uses DU 4.0 on an
AlphaServer 1000.

> - should we be using a more-recent-tho-possibly-broken squid?

1.1.10 is fine. We just upgraded to 1.1.14.

> - anyone have a good recommendation for cpu/mem/swap on a DU box for
> this /some other version of squid?

Three rules with Squid, specially on DU (it's a bit different on Linux):

1) Add RAM
2) Add RAM
3) Add RAM

Squid is a hog. Memory really used depends on the malloc you use and
wether it leaks or not (we found that GNU malloc is a good choice *and*
we restart Squid every week), depends on the size of the memory pools
(cache_mem) and depends on the disk size (I bought more disk than the
memory could use, I now have to beg for RAM).

CPU : any 386 will be more than enough (our cache, with twelve HTTP
requests per second - and more ICP - runs at 30 % of the CPU limit).

Swap: do not even think about it. Response time will dramatically drop
when Squid swaps (I tried!).
Received on Thu Jul 17 1997 - 00:53:31 MDT