Newsgroups: comp.parallel
From: bjl@eecg.toronto.edu (Brian Jonathan Lee)
Subject: Interesting Proposal for NUMA machines (BBN Butterfly and IBM RP3)?
Organization: Computer Group, Elec. Eng., University of Toronto
Date: 16 Sep 1997 13:32:49 GMT
Message-ID: <5vm1q1$j5p$1@server2.ctc.com>

Hello,

In writing up my thesis, I came accross an idea that I thought I'd run
by the ca and cp newsgroups.

In machines such as the BBN Butterfly and IBM RP3, the distributed
memory module sits right beside the processor.  In order to implement
fairness (equal bandwidth, equal accessibility) to all synchronization
objects in memory, we would either need to do some silly things in
software or have all processors to access the memory through the
network.  The local processor wouldn't usually use the network for a
local access and would have to be forced to do so.

Any comments?  Do any machines do this already?

regards,
bjl

Brian Jonathan Lee     ENG SCI 9T4 | "You are the answer, you are my key, 
Computer Group - ECE Department    |  But it is only impossible, for you to
University of Toronto              |    return to me."  -- bjl '90
bjl@eecg.toronto.edu               | http://ncmud.accessv.com/~forge

--
Articles to parallel@ctc.com (Administrative: bigrigg@ctc.com)
Archive: http://www.hensa.ac.uk/parallel/internet/usenet/comp.parallel