Re: Memory requirements, 3GB cache

"Miguel A.L. Paraz" <map@iphil.net> wrote:
>I figure that with a slow link, swapping out to disk when running
>out of physical RAM is not a problem. The machine is currently

>But don't quote me on it, I'm not an OS guru by any stretch of the
>imagination. ;)

Ok, so I quoted you anyway... But just to make a point:

The problems I'm thinking of can occur even if you're on a slow
link. Say, for every request squid gets, it cycles through memory
touching/checking all stored objects. Now, this could mean that
*every* time someone makes a request, it has to have the whole process
in memory. This means that, if the process is too big, it will have
to swap all of squid in and out of memory on every request.

No matter how slow your link is, you'll *always* know the difference
in that case between swapping and not swapping.

Then again, maybe the access patterns for the stored objects are
"swap-friendly", and only a small subset of the stored objects are being
referenced on every request. That would be ideal, but the question is
if squid does it this way. I'm currently already into a 400MB-squid
and 2600MB-spinner ratio (it has been moving data between the two caches
since noon), and would like to know beforehand if I'm heading for disaster.
I have 32MB of RAM, and would hope that squid can serve a 3GB cache with
a resident set of maximally 22MB.

-- 
Sincerely,                                                          srb@cuci.nl
           Stephen R. van den Berg (AKA BuGless).
"I have a *cunning* plan!"