Re: [squid-users] RAID, 64bit and cache_dir size

On Thu, 12 Jan 2006, Gregori Parker wrote:

> I have read that RAID is a bad idea for squid caches

For Internet proxies yes, as they rotate their cache content quite
frequently, causing a lot of small writes which most RAID setups isn't
very happy about.

For accelerators where the content is more stable in nature RAID only
provides benefits.

> however I am unable to find any reasoning for this aside from
> performance concerns.

It's mostly performance concerns.

Exception: RAID0 (concat/striping) is disliked as it does not buy anything
for Squid, only makes your whole cache more vulnerable should one of the
disks in the set fail (whole cache lost instead of just the content of
the failed drive).

> Has anyone had any issues running Squid in a 64-bit environment? I
> plan to use Fedora Core 4 x86_64 and was wonding if anyone had any
> experiences (good or bad) with this.

Squid isn't very much tested running in 64-bit mode in a 64-bit
environments. Running in 32-bit mode should be fully tested however.

Additionally the memory consumption is considerably larger than the
compared setup in a 32-bit environment.

The main benefit of running Squid in 64-bit mode is that it allows for a
very large cache_mem, which is not possible in 32-bit mode due to hardware
constraints on process size.

> Finally, I'm interested in what was just asked about large
> cache_dir's: Is it better to have one large cache_dir (1 TB for example)
> or multiple smaller cache_dir's (5 x 200 GB) - I'm mostly concerned with
> performance and number of file descriptors.

No difference.

> Each server will have 4 GB of RAM, which according to my math, should
> be plenty for this large of a cache...also worth noting that cached
> objects will be a minimum of around 500KB each.

There is a formula in the FAQ giving you an estimate of the index usage
per number of objects. Yes, 4GB will be plenty and more for your setup.

Regards
Henrik
Received on Fri Jan 13 2006 - 04:38:23 MST