Re: Is Cache object real data kept both in memory & disk ?

From: Scott Hess <[email protected]>
Date: Wed, 5 Jan 2000 07:50:32 -0800

Jan Van Ham <Responsible@g-net.be> wrote:
> I don't doubt this BUT when you don't specify a 'higher' cache_mem value
the amount
> of hot-object will ALWAYS be low (as reported in cachemanager
....offcourse I don't
> know wether it tells the truth on this one)
>
> I've tested with cache_mem 64MB and the amount of mem-objects went
> up up up up! (more then 10k) (with the default of 8MB it always is only
> about 1500) So I don't know what to think about it ...

But did your performance improve on any front? Serving objects from
Squid's memory is only a gain if they actually get served faster...

Keep in mind that Squid has issues with swapping. Since it's a
select-based server, if there's an in-memory object that got swapped out,
if _anyone_ requests that object _all_ clients will hang waiting for it to
be swapped back in (Squid can't service other clients while it swaps back
in). The more cache_mem you use, the greater your chance of getting a page
swapped out.

This is generally less of an issue for dedicated Squid boxes, because you
don't have to worry about other processes forcing Squid to swap out. Also,
it's not horribly bad if it swaps infrequently - though keep in mind that
some operating systems (default for Linux, I think) may swap virtual memory
pages out in order to maintain enough disk buffers, meaning that on
high-volume systems, Squid's own activity can swap Squid out.

Later,
scott
Received on Wed Jan 05 2000 - 09:05:16 MST

This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 16:50:14 MST