From: nmm1@cus.cam.ac.uk (Nick Maclaren)
Newsgroups: comp.parallel.mpi
Subject: Re: Performance of overlapped communication and computation
Date: 30 Jan 1999 10:46:25 GMT
Organization: University of Cambridge, England
Message-Id: <78unu1$mta$1@pegasus.csx.cam.ac.uk>
References: <78sjcr$ath$1@lux.doc.ic.ac.uk> <78tfeo$qsr@news3.newsguy.com>
Xref: ukc comp.parallel.mpi:4568


In article <78tfeo$qsr@news3.newsguy.com>,
Greg Lindahl <lindahl@cs.virginia.edu> wrote:
>wjk@pluto.doc.ic.ac.uk (William Knottenbelt) writes:
>
>> I had a quite a lengthy reply from Fujitsu today detailing why the
>> AP3000 can't overlap communication and computation like the IBM SP-2
>> does. The main problem seems to be the lack of direct user level
>> access to the underlying network hardware (Solaris would have to be
>> modified to allow this).
>
>You can do this with just a device driver -- Myrinet, for example,
>does this correctly on Slowaris.

But, as been repeatedly pointed out, that is NOT the main problem.