From chemistry-request -x- at -x- server.ccl.net Wed Jun 12 03:28:56 2002
Received: from ns.ice.mpg.de ([195.37.47.10])
	by server.ccl.net (8.11.6/8.11.0) with ESMTP id g5C7Suj10964
	for <chemistry # - at - # ccl.net>; Wed, 12 Jun 2002 03:28:56 -0400
Received: from relax.ice.mpg.de ([10.10.11.61] helo=ice.mpg.de)
	by ns.ice.mpg.de with esmtp (Exim 3.14 #17)
	id 17I2Yp-0005vq-00; Wed, 12 Jun 2002 09:28:55 +0200
Message-ID: <3D06F836.6010103: at :ice.mpg.de>
Date: Wed, 12 Jun 2002 09:28:54 +0200
From: Christoph Steinbeck <steinbeck $#at#$ ice.mpg.de>
Reply-To: steinbeck(+ at +)ice.mpg.de
User-Agent: Mozilla/5.0 (Windows; U; Windows NT 5.0; en-US; rv:1.0.0) Gecko/20020530
X-Accept-Language: en-us, en
MIME-Version: 1.0
To: FyD <fyd: at :u-picardie.fr>
CC: chemistry %-% at %-% ccl.net, Gil.Utard %-% at %-% ens-lyon.fr
Subject: Re: CCL:Cluster for AMBER/GAMESS/GAUSSIAN
References: <20020611144513.5829F7EB1&$at$&gip.u-picardie.fr>
Content-Type: text/plain; charset=us-ascii; format=flowed
Content-Transfer-Encoding: 7bit

FyD wrote:
> Dear All,
> 
> We are going to buy a cluster of dual athlon MP 2000+ to work with AMBER6/7,
> GAMESS and GAUSSIAN.
> We hesitate between - a Myrinet 2GB or a 1GB Ethernet network and 
>                     - SCSI HDs 15000 rpm/SCSI 10000 rpm or IDE 7200 rpm
> 
> Both Myrinet 2GB and SCSI HD 15000 rpm are really more expensive but might
> or might not be crucial...
> Moreover, GAMESS, GAUSSIAN and AMBER might not need the same network or HDs
> for optimizing calculations...
> 
> What is crucial for AMBER (MM) and for GAMESS/GAUSSIAN (QM) ?
> 
> Thanks to all, Kind regards, Francois

My 16-Nodes linux cluster (http://www/departments/ChemInf/edda.html) 
runs on 100 MBit Ethernet. The Gaussian 98 calculations scale nicely 
(http://www/departments/ChemInf/g98scaling.html) for up to 8 nodes. 
Please note, that this might be true only for the B3LYP geometry 
optimization that we do.

Due to this behaviour, I have logically splitted the cluster into two, 
using the PBS batch system, because we run a lot of jobs and two 
simultaneous jobs on 8 nodes each are much more sensible then one on 16, 
if you have hundreds of jobs in the queue.

For us, the 100 MBit Ethernet seems to be ok. We have been convinved by 
others that ab-initio calculations of the type that we run always show 
that scaling behaviour that we see and that thus a faster network would 
be a waste of money. I understand that today GBit Ethernet costs as must 
as did the 100 MBit stuff at the time we bought our cluster, so this 
would be my choice.

I have nothing to contribute with respect to HD speed and MM calcs :-)

Cheers,

Chris

-- 
Dr. Christoph Steinbeck (http://www.ice.mpg.de/departments/ChemInf)
MPI of Chemical Ecology, Winzerlaer Str. 10, Beutenberg Campus, 07745 
Jena, Germany
Tel: +49(0)3641 571263 - Fax: +49(0)3641 571202

What is man but that lofty spirit - that sense of enterprise.
... Kirk, "I, Mudd," stardate 4513.3..