On Sun, 11 Apr 2004, Mark Tinka wrote:

> what happenned was that my /swap was on and running, but my system never
> ever did swap, even when the load increased.. physical RAM utilisation
> went from 1.3GB to as low as 2MB.. whenever RAM hit 2MB, it shot up to
> 3MB, and this cycle continued until either Squid did its internal
> sanities, or i restarted the Squid process, at which point memory was
> freed up to about 24MB.. and then continued to fall till 2MB, where it
> rose back again to 3MB, and this went on and on and on and on...

Then your sustem never reached the point that there was memory pressure,
i.e. you had significantly more than sufficient amount of memory.

> i would like to achieve this, but the difference is that i will be
> running on a 2.4 kernel, and not a 2.2, as is the previous Squid server,
> so am not sure how good my chances are.. ..

Test it.

> is there a clear explanation as to why this is..?..

It is hard to give clear explanations on VM behaviour. It is a complex
business with very many variables.

But what can be said is that it is good to reserve a significant chunk of
memory for the OS maintenance of filesystems buffers etc.. The Squid rule
of thumb generalises this a bit and gives the same amount as the actual
Squid usage to OS and buffers (ca twice the memory than used by the Squid
process). For most Squid setups this balances well allowing most VM
implementations to do it's job without disturbing Squid much..