[[ isn't this a netbsd-users question? ]]

[ On Thursday, December 5, 2002 at 16:18:07 (+0100), Bernd Sieker wrote: ]
> Subject: Re: increasing a process's max data size
>
> Otherwise I would rather restrict squid's memory usage. (It makes
> no sense to have squid cache more data in "RAM" then you have actual
> RAM.  squid itself is probably better in writing unused data to
> disk than the VM system (for this particular task.)

Yes, but you can't reliably restrict squid's memory usage from the
outside -- at least not with some of the versions (I've not yet tested
2.5) as that'll just make it think somethings wrong and will _usually_
cause it to crash.

To control squid's virtual size you have to control it from the inside
-- i.e. using its own internal configuration parameters.

> Note that the actual squid process may become a lot bigger than
> what you set in the cache_mem variable . I use cache_mem 16 MB,
> and the actual process size is usually around 40-45 MB. This is a
> squid for a small LAN (10 machines sharing one ADSL 768/128 line)

Much of Squid's memory usage is for all intents and purposes
"controlled" by the number of on-disk objects it's allowed to keep.  The
data structures representing these objects are all kept in in memory.

The bigger your cache disk(s) (and the smaller your average objects)
then the more memory the main process will consume and the more trivial
a "reasonable" setting of cache_mem becomes to the entire usage picture.

-- 
								Greg A. Woods

+1 416 218-0098;            <g.a.woods@ieee.org>;           <woods@robohack.ca>
Planix, Inc. <woods@planix.com>; VE3TCP; Secrets of the Weird <woods@weird.com>