RE: [squid-users] Squid Memory Limitations?

From: Andrew Sawyers <[email protected]>
Date: Thu, 21 Nov 2002 11:33:11 -0500

Ah, you're right.....I'd forgotten about this behavior.
ATM we have about 300MB of free ram....so this is fine for our systems. The
obvious main concern was the squid process itself exceeding 2048MB.
Thanks,
Andrew

-----Original Message-----
From: Joe Cooper [mailto:joe@swelltech.com]
Sent: Thursday, November 21, 2002 11:21 AM
To: Andrew Sawyers
Cc: Robert Collins; squid-users@squid-cache.org
Subject: Re: [squid-users] Squid Memory Limitations?

Hmmm...I would expect the OS to fill that extra memory with cached disk
data, eventually. Assuming you've got enough disk activity to fill it.

What does top tell you about buffer and cache usage? If you have a
large enough data set, 'free' should drop down to just a few MB, while
buffer and cache should be quite large (in the hundreds of MB). Which
means you are taking advantage of the extra RAM, even if Squid isn't
using it directly--when Squid wants something from disk, if the OS has
it cached, it will still come from RAM. Not quite as fast as Squid
accessing it directly from its own memory pools, as there is the extra
overhead of a filesystem request and some extra copying, but certainly
faster than coming from disk.

Andrew Sawyers wrote:

> Well, the os isn't leaving it all unused; thus why I said ~2GB. These are
> only doing cacheing, so outside of whatever ram useage squid is using and
> normal operating system stuff, the rest will be unused, no?
> Andrew
>
> -----Original Message-----
> From: Joe Cooper [mailto:joe@swelltech.com]
> Sent: Thursday, November 21, 2002 10:58 AM
> To: Andrew Sawyers
> Cc: Robert Collins; squid-users@squid-cache.org
> Subject: Re: [squid-users] Squid Memory Limitations?
>
>
> Andrew Sawyers wrote:
>
>
> >Robert,
> >I was asking because one of our squid accelerated cache servers has
> 4GB of
> >RAM in it and we saw a scenario in which it looks like squid restarted
> >suspiciously close to 2048MB in process size.
>
>
> ...
>
>
> >Since we noticed this suspicious restart on the biggest cache server
> >first,
> >I thought I'd find out for sure if there could be a problem there. We've
> >set back our cache_mem to ~550MBs so we don't expect to exceed 2GB
> >again...we're just leaving ~2GB of ram unused by doing so.
>
>
> If your OS leaves this memory unused, you need a better OS. ;-)
>
> The OS buffer/cache ought to do good things with that spare memory...In
> my experience, after about 64MB or so, doubling cache_mem does almost
> nothing for performance. And yet halving system mem from 512 to 256
> makes a big difference. Scale this to your case, and I suspect you
> might still see a drop in performance if you removed the extra 2GB.
> Just a guess, though.

--
Joe Cooper <joe@swelltech.com>
Web caching appliances and support.
http://www.swelltech.com
Received on Thu Nov 21 2002 - 09:33:32 MST

This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 17:11:21 MST