[KLUG Members] Re: Clustering -- depends on the app

Bryan J. Smith members@kalamazoolinux.org
Thu, 10 Jan 2002 13:07:34 -0500


Dan Downs wrote:
>    Part 1.1    Type: Plain Text (text/plain)
>            Encoding: quoted-printable
> A friend of mine is looking into some powerful servers
> ($50K-250K) and asked for some recommendations. I suggested
> clustering since they'll be using linux, but I've got some
> questions about it.
> Is clustering transparent to the apps?

Depend on the app.  See below.

> Do applications have to be programmed to support it?

Depends on the app.  See below.

> Has anyone here used clustering? If yes any info on setup/
> distributions?

Yes, for various apps.

For engineering, it depends on the app.  Most statistical, scientific
and engineering apps that use multiple, independent runs work great! 
I've done this for Monte Carlo methods, Computation Fluid Dynamics
(CFD), various "batch runs" inputted/scheduled from different users,
etc...  My "domain" of clustering has been for engineering purposes,
usually $15-50K clusters of 10-50 systems.

If you are looking for "high availability" file/web servers, there are
various avenues.  For web servers, it could be as simple as using
2.4/Netfiler at your firewall, and "load-balancing" incoming requests
(largely read-only access) across a couple of regular IPs.  For
file/database (or more "read/write" access), it gets tricky.  Mission
Critical Linux specializes in fail-over NFS (and Samba) servers, there
is AFS, Coda and a number of [increasingly] new "distributed
filesystems."

After getting laid off last April, I almost got a job with Mission
Critical Linux (and Cray), and played with some of their NFS stuff (only
to see both of them go on a "hiring freeze" in May due to the economy,
admist interviews ;-).

Again, it all depends on the application.

-- Bryan

-- 
Bryan J. Smith, Engineer          mailto:b.j.smith@ieee.org
AbsoluteValue Systems, Inc.       http://www.linux-wlan.org
SmithConcepts, Inc.            http://www.SmithConcepts.com