[Bioclusters] Request for discussions-How to build a biocluster Part 3 (the OS)

[Donald Becker]

On 9 May 2002, Mike Coleman wrote:

> Sylvain Foisy <foisys@mac.com> writes:
> >
> > Well, it will be Linux of course. Around here, we use Red Hat 7.2, Mandrake
> > and Debian. Any preferences? Any one of them is easier to install/configure
> > for clusters?
>
> I'm a big Debian fan, so I'm looking at a way to use it.  One annoyance I see
> with various bits of prepackaged clustering systems/software is that *they*
> want to choose the distribution for you (usually they choose Red Hat).  This
> seems a bit like the tail wagging the dog to me.

Not at all.
You selection will depend on the type of cluster you want, and how much
effort and expertise you will need to build and maintain the cluster.

You can build an first generation Beowulf cluster by putting together a
collection of independent machine running your favorite distribution.
That will require loading the distribution on each machine, and careful
control of system and application updates.  It's suitable for one person
being the combined builder, administrator and user.

The Scyld distribution includes features such as unified process space,
single point/ single binary application installation, cluster directory
services and zero-administration compute nodes.  These features require
kernel and library support.

I'm trying avoiding making this too much of a marketing post, but there
is a big usability difference between an architected cluster system and
a collection of machines.

> Well, the nodes could definitely nfs-boot from a single master whether or not
> they actually have local disks.

Doing diskless boots over NFS works for a handful of nodes, but is a
significant performance and scalability bottleneck.  There are ways to
mitigate the performance problems with tuning, but you have to measure
and monitor the system to verify that it's performing as you expect.

> ...The idea is that they would get sucked down once after boot
> and basically sit in RAM.

You can either copy to a local disk, or set the NFS attribute and data
cache expiration to a large number.  Both require understand the
synchronization and update semantics of NFS.

-- 
Donald Becker				becker@scyld.com
Scyld Computing Corporation		http://www.scyld.com
410 Severn Ave. Suite 210		Second Generation Beowulf Clusters
Annapolis MD 21403			410-990-9993