From owner-nwchem-users  Mon Mar 31 17:25:11 2003
Received: (from majordom@localhost)
	by odyssey.emsl.pnl.gov (8.8.8+Sun/8.8.5) id RAA18630
	for nwchem-users-outgoing; Mon, 31 Mar 2003 17:08:48 -0800 (PST)
Date: Mon, 31 Mar 2003 17:06:59 -0800
From: Richard Massa <ramassa@ucdavis.edu>
Subject: Re: nwchem diskless compute nodes?
In-reply-to: <"from webster"@mail.chem.tamu.edu>
To: Charles Edwin Webster <webster@mail.chem.tamu.edu>
Cc: nwchem-users@emsl.pnl.gov
Mail-followup-to: Richard Massa <ramassa@ucdavis.edu>,
 Charles Edwin Webster <webster@mail.chem.tamu.edu>, nwchem-users@emsl.pnl.gov
Message-id: <20030331170659.P23062@tictactoe.unixboxen.net>
MIME-version: 1.0
Content-type: text/plain; charset=us-ascii
Content-disposition: inline
User-Agent: Mutt/1.2.5i
References: <3E88D9AC.FF63A8E9@mail.chem.tamu.edu>
Sender: owner-nwchem-users@emsl.pnl.gov
Precedence: bulk

Of course, disks in nodes are faster, but diskless nodes are easier to manage.
If you were to do this, I'd set up a parallel fast or gigabit ethernet network
to run the NFS connections, because myrinet has very poor nfs performance, and
a fast ethernet network is so cheap to set up.

On Mon, Mar 31, 2003 at 06:13:33PM -0600, Charles Edwin Webster wrote:
> all,
> 
> has anyone had experience running nwchem on diskless
> compute nodes on a AMD or Intel Cluster with SCSI RAID?
> 
> the interconnect would likely be myrinet or infiniband
> to keep the latency for IO to a minimum.
> 
> would the speedup from myrinet be negated by non-local
> scratch?
> 
> my feeling is that this would not be as efficient as local
> SCSI drives, but we need to have a definite answer.
> 
> thanks in advance,
> 
> charles webster
> 

-- 
"Sonny Bono Copyright Term Extension Act... The meaning is: No one can do to
the Disney Corporation what Walt Disney did to the Brothers Grimm."
													-Lawrence Lessig