From owner-nwchem-users@emsl.pnl.gov Fri Aug 13 20:41:03 2004
Received: from odyssey.emsl.pnl.gov (localhost [127.0.0.1])
	by odyssey.emsl.pnl.gov (8.12.10/8.12.10) with ESMTP id i7E3f2KQ028601
	for <nwchem-users-outgoing@odyssey.emsl.pnl.gov>; Fri, 13 Aug 2004 20:41:02 -0700 (PDT)
Received: (from majordom@localhost)
	by odyssey.emsl.pnl.gov (8.12.10/8.12.10/Submit) id i7E3f2bb028600
	for nwchem-users-outgoing; Fri, 13 Aug 2004 20:41:02 -0700 (PDT)
Date: Fri, 13 Aug 2004 20:40:59 -0700
From: "Nieplocha, Jarek" <jarek.nieplocha@pnl.gov>
Subject: Re: NWChem-4.6 &Infiniband from Mellanox
To: kus@free.net
Cc: nwchem-users@emsl.pnl.gov
Message-id: <A037CF41DB4D984C9A3C3024C612666E06F41F@pnlmse25.pnl.gov>
MIME-version: 1.0
X-MIMEOLE: Produced By Microsoft Exchange V6.5.6944.0
Content-type: text/plain; charset=US-ASCII
Thread-Topic: NWChem-4.6 &Infiniband from Mellanox
thread-index: AcSBWkMM9gPJGP8oT6moh59immJuRQAVkB14
Content-class: urn:content-classes:message
X-MS-Has-Attach: 
X-MS-TNEF-Correlator: 
X-OriginalArrivalTime: 14 Aug 2004 03:41:00.0601 (UTC)
 FILETIME=[843CA290:01C481B0]
Content-Transfer-Encoding: 8bit
X-MIME-Autoconverted: from quoted-printable to 8bit by odyssey.emsl.pnl.gov id i7E3f0KQ028597
Sender: owner-nwchem-users@emsl.pnl.gov
Precedence: bulk


Nwchem runs over tcgmsg-mpi on top of MPI.

Regarding LAM MPI support for IBA, LA MPI and LAM MPI are merging. There is a version developed at Sandia that supports IBA. Not sure if this code has been officially released yet but it is available. 


(Sent from my wireless RIM Blackberry PDA)


-----Original Message-----
From: Mikhail Kuzminsky <kus@free.net>
To: Nieplocha, Jarek <jarek.nieplocha@pnl.gov>
CC: nwchem-users@emsl.pnl.gov <nwchem-users@emsl.pnl.gov>
Sent: Fri Aug 13 10:23:31 2004
Subject: Re: NWChem-4.6 &Infiniband from Mellanox

In message from "Nieplocha, Jarek" <jarek.nieplocha@pnl.gov> (Fri, 13 
Aug 2004 09:03:19 -0700):
>In principle, you are correct about Item 2. 
   Thanks for your reply ! Does it means that if I'll use in NWChem 
own diagonalization routines w/direct MPI calls instead of PEIGS,
I'll have no needs in GA/ARMCI and therefore will work only w/"pure 
MPI" ?
  
>We did not test NCSA MPI. 
  No problems.
>There is also LAM MPI over IBA to consider.
  ? AFAIK, LAM MPI 7.0.6 don't have native IB support. May be it's 
possible to use TCP/IP w/IPoIB, but I don't know - will it works 
w/o things like SDP or no :-(

Yours
Mikhail

>
>-----Original Message-----
>From: owner-nwchem-developers@emsl.pnl.gov 
><owner-nwchem-developers@emsl.pnl.gov>
>To: nwchem-users@emsl.pnl.gov <nwchem-users@emsl.pnl.gov>
>Sent: Fri Aug 13 06:06:06 2004
>Subject: NWChem-4.6 &Infiniband from Mellanox
>
>1) I'm considering the inst of NWChem 4.6 on Opteron/Infiniband Linux 
>cluster (SuSE Linux 9.0, kernel 2.4.21 for x86_64, smp).
>IB equipment is from Mellanox, so ARMCI-1.1 must work because it's
>based on Mellanox VAPI.
>
>OK, but what is about MPI over IB ? It looks for me that both free 
>MPI 
>realizations (from NCSA and from Ohio) for IB will work - is it right 
>?
>
>
>2) Pls confirm or reject my "understanding":
>
>a) I may use any MPI realization instead of TCGMSG b) Dependence from 
>interconnect hardware is present only for Global Arrays which is used 
>only in PEIGS library. Global Arrays is
>used in NWChem *in addition* to direct MPI calls.
>
>c) To be more exactly, GA uses ARMCI which is only 
>"interconnect-depended" software component of NWChem. ARMCI gives to 
>programmer some own simple "API" for one-sided communications which 
>are based on low level software (not on MPI).
>
>Yours
>Mikhail Kuzmisnky
>Zelinsky Institute of Organic Chemistry
>Moscow
>  
>
>
>
>