RE: [squid-users] Squid Caches

From: Robert Collins <robert.collins@dont-contact.us>
Date: Mon, 30 Apr 2001 09:50:10 +1000

> -----Original Message-----
> From: wojtek@3miasto.net [mailto:wojtek@3miasto.net]
> Sent: Sunday, April 29, 2001 8:34 PM
> To: Robert Collins
> Cc: Mike Diggins; squid-users@squid-cache.org
> Subject: RE: [squid-users] Squid Caches
>
>
> > Subject: RE: [squid-users] Squid Caches
> >
> > http://www.squid-cache.org/mail-archive/squid-users/200008/0147.html
>
>
> > Should cover the perf reasons. (in summary raid 5 has read
> perf benfits
> > only, and can affect squid (depending on your hit rate vs
> object commit
> > rate of course).
>
>
> read perf could be better for one-by-one request. but if we
> use diskd so
> doing many requests on many cache dirs in paralell it would be fastest

Did you read the link I provided? The question was "why is striping not
as good as multiple cache dirs (1 per disk) ".

As for using diskd you are mostly correct -but-
- for diskd the best is achieved with more than one cache dir per
physical disk, no OS or hardware striping. This is because diskd doesn't
make use of O/S / hardware parallelism (ie only one request outstanding
at a time)
- for aufs one cache dir per physical disk (aufs can handle multiple
concurrent requests per cache dir).

There are some documents on the suqid-cache web site describing the
config for one of the cache bakeoffs - very interesting reading.

Rob
  
> > See the raid branch at sourcerforge for auto-recovery after
> cache_dir
> > failures.
> >
> > Rob
> >
> > > -----Original Message-----
> > > From: Mike Diggins [mailto:diggins@mcmail.cis.mcmaster.ca]
> > > Sent: Sunday, April 29, 2001 11:59 AM
> > > To: squid-users@squid-cache.org
> > > Subject: Re: [squid-users] Squid Caches
> > >
> > >
> > >
> > > Thanks for the replies.
> > >
> > > My initial thought was that a stripe set was the obvious
> > > choice for the
> > > squid cache. It seems that isn't the case. Why wouldn't Squid
> > > benefit from
> > > the stripe set?
> > >
> > > Also, if I have, say, two cache dirs spread over two physical
> > > disks and
> > > one drive fails, will squid continue to run with one
> cache dir out of
> > > commission?
> > >
> > > -Mike
> > >
> > > On Sat, 28 Apr 2001, Henrik Nordstrom wrote:
> > >
> > > > There is no big performance difference between the two
> > > configurations,
> > > > but having them separate is probably easier to maintain.
> > > >
> > > > Note: If you have disksuite then there is a performance gain in
> > > > assigning one of the drives as DiskSuite log drive for the other
> > > > drives.. (the log drive should not be used for cache, but
> > > it might be
> > > > used for OS or other less I/O intensive purposes)
> > > >
> > > > --
> > > > Henrik Nordstrom
> > > > Squid Hacker
> > > >
> > > > Mike Diggins wrote:
> > > > >
> > > > > What's the better option for a squid cache; multiple
> > > cache directories
> > > > > over two or more separate disks or a single cache on a
> > > striped disk set
> > > > > (say two or three disks)? This would be on a Sun system
> > > running solaris 8
> > > > > and squid 2.4 stable1.
> > > > >
> > > > > -Mike
> > > >
> > > >
> > >
> > >
> >
>
> -----------------------------------------------
> UNIX *is* user friendly.
> It is just a bit selective about his friends.
>
>
Received on Sun Apr 29 2001 - 17:58:02 MDT

This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 16:59:40 MST