RE: [squid-users] squid and kswapd cpu problem

From: Anders Larsson <anders@dont-contact.us>
Date: Mon, 20 Oct 2003 16:25:36 +0200

Should't this be relative to the amount of clients that connects to the
machine ?
The machine as im using serves 5k users
therefore i thought it should have a larger cache_mem.

the Squid proccses at the moment RSS 64Mb
ill check later and see it's get any higher.

// Anders

On Mon, 2003-10-20 at 16:06, Adam Aube wrote:
> > But the machine has 4Gb ram so why not using 500MB ?
> > is it better to let squid increase if it's needed?
>
> Linux will aggressively use free RAM to cache files. If Linux does
> most of the caching, it benefits the entire system, not just Squid,
> and Linux will dynamically adjust the amount of memory used to cache
> files based on current system memory needs, whereas Squid will always
> try to completely fill the memory alloted to it by cache_mem.
>
> Additionally, I don't think Squid will have any way of knowing if the
> memory allocated to it for cache_mem is swap and not RAM (perhaps
> Henrik or Robert can comment on this).
>
> That said, it is good to guarantee a small amount of memory for
> caching very "hot" objects; that is what cache_mem is for, and that is
> why it defaults to 8 MB.
>
> Adam

-- 
Received on Mon Oct 20 2003 - 08:37:16 MDT

This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 17:20:33 MST