hardware nuts & bolts

From: Eugen Leitl (eugen@leitl.org)
Date: Thu May 30 2002 - 14:22:28 MDT


According to news from the trenches, 1 kNodes nowadays is not considered
large anymore. There is a distinct trend to more dense packaging, and
advent of first fully solid state clusters. Custom high-speed
interconnects are getting better, but Ethernet and GBit Ethernet appears
to be dominating for price/performance for budget clusters.

This trend should hold for the next half decade, with advent of 64 bit low
end and built-in message passing, plus multi-CPU modules.

---------- Forwarded message ----------
Date: Wed, 29 May 2002 07:52:35 -0600 (MDT)
From: Ronald G Minnich <rminnich@lanl.gov>
To: John Hearns <john.hearns@cern.ch>
Cc: linuxbios@lanl.gov
Subject: Re: Maybe stupoid question...

On 29 May 2002, John Hearns wrote:

> When installing a fresh node in a cluster,
> we would like it the first time to boot over the
> network. The second time it boots, it should boot from
> its own hard disk.

I don't know about this local disk thing. The next cluster we build (1024
node) won't have 'em. Cplant@sandia.gov, an 1824-node cluster, doesn't
have them.

What do you use local disk for?

> Also, if we started to use Linuxbios, is there any way
> Linuxbios could help with this task.

our next, 1024-node cluster, will be linuxbios-based. As we learned on our
128-node Alpha cluster, linuxbios helps a whole lot ...

ron



This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:39 MDT