[Beowulf] SGI to offer Windows on clusters

[Michael Will]

<disclaimer>
I work for Penguin and it's my job to sell Scyld, however the following
also represents my personal
opinion.
</disclaimer>

The issue you mentioned below is why I love the light weight compute
node concept of Scyld,
which installs on top of redhat enterprise 4. 

While it does still install everything that redhat 4 would install on
the headnode that you interact with,
the compute nodes will not get any of that. 

The compute nodes just boot precisely what is needed, the kernel, the
modules for the hardware detected,
and only when an app is migrated out at runtime it will pull over
additional required libraries, caching
them for next time. It boots directly into RAM (no nfs-root overhead). I
just ran 'free' on a compute node
of one of our demo clusters that has been up for several months and it
still only uses 140MB of it's 2G - 
the only process running on it currently is portmap since it has /home
nfs-mounted.

The downside is that you have to configure the environment for specific
applications that expect everything 
to be there, i.e. if they do system("/usr/bin/perl") and runtime and
what-not, that is when you start giving up some
of the pure principle and solve it by i.e. mounting /usr via NFS or
predeploying out stuff. 

Michael Will
SE Technical Lead
Penguin Computing Inc.
-----Original Message-----
From: beowulf-bounces at beowulf.org [mailto:beowulf-bounces at beowulf.org]
On Behalf Of Joe Landman
Sent: Monday, April 16, 2007 8:40 AM
To: Robert G. Brown
Cc: beowulf at beowulf.org
Subject: Re: [Beowulf] SGI to offer Windows on clusters



Robert G. Brown wrote:

> I've always liked the idea of the core remaining a VERY marginal set 
> that is pretty much "just enough" to bootstrap an install.  One of the

Hmmm.... I indicated this some time ago and got some grief over this. 
Few of the distro makers seem to like this concept.  I want the unit up,
running all drivers needed (and only those drivers needed), with the
network, and a barebones admin package (ssh, ipmitools, ...).  Getting
there with most prebuilt distros is excruciatingly hard.

I have gotten SuSE down to a 1.4 GB install, bare minimum that I can
make it and satisfy all dependencies, and have a functional compute
node.  More if I need to get the 32 bit packages there.

RH I have not seen an install below about 2GB.  Rocks tries to do a
minimal install, and yet you see packages that largely won't be used on
nodes being installed in order to satisfy dependencies of packages that
will be used.  This is not the Rocks people's fault, it is the
underlying distribution.

You can use DSL and other "small" distros as long as you don't mind
using ancient kernels, missing drivers, ...  This unfortunately doesn't
satisfy the needs.  Caos3 should (in theory, haven't tried it yet)
handle most of this, be a very tight install, limited package dependency
radius, and still be quite usable/fast.

> things that from time immemorial has bugged me about the red hat 
> install process is its complete lack of robustness, so that if for any

> reason it fails in midstream, one pretty much has to start over.  This

> has always

YES!!!!!

These problems plague every RH derived distro as well.  FC derived as
well.  Anaconda may be great at many things.  Robustly installing
software and configuring systems is not one of them.



--
Joseph Landman, Ph.D
Founder and CEO
Scalable Informatics LLC,
email: landman at scalableinformatics.com
web  : http://www.scalableinformatics.com
phone: +1 734 786 8423
fax  : +1 734 786 8452
cell : +1 734 612 4615
_______________________________________________
Beowulf mailing list, Beowulf at beowulf.org To change your subscription
(digest mode or unsubscribe) visit
http://www.beowulf.org/mailman/listinfo/beowulf