On Wed, Sep 19, 2007, Ding Deng wrote:
> > It'd mean more RAM used to track expired objects, and more CPU to walk
> > the list and delete unneeded objects..
>
> And probably longer disk seek time.
Depends how its done. Doing it on a busy UFS might mess with your
service times.
> Agreed. We still have to make sure that cache_dirs match server memory
> however, as I'm seeing a scenario right now that if we make full use of
> all the available disks, a single Squid instance will eat up dozens of
> gigabytes of physical memory, which is far more than what we've
> installed on the server.
Squid isn't exactly memory-frugal at the present time. I've been thinking
of ways to improve its index memory usage.
> I'm also seeing a scenario that 10GB of cache_dirs get roughly the same
> hit ratio as 30GB of cache_dirs due to cache pollution, so cache_dir
> which is larger than necessary is not always a good idea.
Yup! Well, caching has diminishing returns with cache size if you're
just caching small objects.
Adrian
-- - Xenion - http://www.xenion.com.au/ - VPS Hosting - Commercial Squid Support - - $25/pm entry-level bandwidth-capped VPSes available in WA -Received on Tue Sep 18 2007 - 23:35:27 MDT
This archive was generated by hypermail pre-2.1.9 : Mon Oct 01 2007 - 12:00:02 MDT