Re: Memory problems

On Thu 28 Nov, 1996, David Lee {DSL AK} <DavidL@datacom.co.nz> wrote:
>2) Why is it not releasing unneeded memory back to the OS when the
>memory_pools option is set to 'off'

this I can answer: Very few malloc() implementations actually
release memory back to the OS when you free(), until the process
exits.

As to the memory usage, a lot comes down to the in-transit memory
objects that reside in memory until they're complete, and the
cache hits which are also loaded entirely into memory whilst being
retrieved. With slow (eg modem) based readers, you can quite easily
tie up a lot of memory.

My own experience (with two 256MB Challenge S) isn't quite as bad as that
- they typically take a couple of days before they get close to where
they might want to swap. One tip - ditch libmalloc and the libc
implementation, and use GNUmalloc. You lose the mallinfo() bits
though, at least in the vesion that I could find.

[Though in fact at the moment Squid 1.0.20 is crashing regularly on
both systems after about one days uptime (due I think to one receiving
a strange URL request, and then the other one dying when it receives it
as an ICP request from the first. If I get a usable core dump, I'll try
to get the URL that caused it). I've not seen any changes in Squid 1.1b*
that look like they might be related]

James.
Received on Wed Nov 27 1996 - 18:31:21 MST