Re: diskstriping or muliple cachedirs

On Mon, 15 Sep 1997, Oskar Pearson wrote:

> I would suggest striping the disks at an OS level... I tried it the
> other way for a while, and had hassles because the disk partitions were
> never quite the same size... this meant that I would suddenly run out of
> space, even though one of the disks had a few 100 megs free..
> and squid would crash... sigh.

Would seem to depend on differing environments, I guess. We use a Sun
Ultra Enterprise server running Solaris, so obviously when it came time to
create "a whopping big cache" I looked to Solstice DiskSuite for creating
a stripe across multiple disks of identical size/geometry.

Admittedly maybe never having used DiskSuite before had a factor in the
decision (what, read the "reference" before the "user guide" - who the
hell wrote this doco anyway?!?!), but I feel reasonably comfortable
having gone down the "multiple cache dir" path (for now, anyway!):

- less software on the machine, lower administrative/operational overhead
  and complexity;

- no need to "waste" a full fast disk on the suggested 100Mb log, or
  to rely on a slow disk for the same transaction log;

- easier to recover from the loss of one disk/dir than when a disk in the
  middle of a RAID0 stripe goes bad; and,

- support for multiple directories already in Squid, so why add extra
  software to duplicate the functionality?

Maybe something like Linux has a much smaller/easier RAID0 striping
environment which would make that choice more attractive, but in our case
there didn't seem to be much point.

Our proxy server handles (according to "calamaris") up to (and beyond)
126,000 requests per hour (combined HTTP/ICP - >60% of this is ICP) and
does nothing else (ie., it's just a proxy server). Responses are fast,
the machine isn't heavily loaded and "all's well".

Using a stripe does seem to often "make sense" and it's an attractive
option, but before doing so have a careful think about your individual
environment and needs - for us, using multiple cache directories made more
sense and so far it's been perfectly fine.

Cheers..

David J. N. Begley <d.begley@nepean.uws.edu.au>
Network Analyst, Communications Unit
University of Western Sydney, Nepean
Australia
Received on Mon Sep 15 1997 - 19:31:27 MDT