Message-Id: <36B96652.6884D202@dolphinics.com>
Date: Thu, 04 Feb 1999 09:20:18 +0000
From: James Cownie <jcownie@dolphinics.com>
Organization: Dolphin Interconnect Solutions
Mime-Version: 1.0
Newsgroups: comp.parallel.mpi
Subject: Re: Performance of overlapped communication and computation
References: <76tkr7$efc$1@lux.doc.ic.ac.uk>
    <772v8p$8q7$1@nnrp1.dejanews.com>
    <3695C224.4557861A@mufasa.informatik.uni-mannheim.de>
    <77505v$749$1@lux.doc.ic.ac.uk> <775lro$t00$1@walter-fddi.cray.com>
    <7767ip$qq7$1@lux.doc.ic.ac.uk> <36974F60.59E2@radia1.com>
    <777qun$i35$2@lux.doc.ic.ac.uk> <3697763F.3F54@radia1.com>
    <78sjcr$ath$1@lux.doc.ic.ac.uk>
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
Xref: ukc comp.parallel.mpi:4583


William Knottenbelt wrote:
> Matt Nichols of Fujitsu Systems (Europe) wrote :
> > I'm not aware of any other vendors who support user-level access,
> > except for maybe Cray/SGI on their vector machines, but again that
> > would be using a proprietary version of UNIX.

Of course direct access to comms hardware from multiple user processes 
in a full Solaris/SPARC environment was available on the Meiko CS2
machine about 5 or six years ago. 

The CS2 machine that LLNL had has just been de-commissioned, so this
isn't new stuff !

The latest revision of that technology is now available from Quadrics 
(www.quadrics.com) for SPARC/Solaris or Compaq Alpha based machines 
(running Tru64 Unix [Digital Unix !]), or is also sold by Compaq
themselves.

A big 21264 based machine coupled up like that ought to be pretty
impressive.
I believe their MPI point to point zero length latency is well below
10uS.

-- Jim 

James Cownie	<jcownie@etnus.com>
Etnus, Inc.     +44 117 9071438
http://www.etnus.com