Newsgroups: comp.parallel
From: moorej@chert.CS.ORST.EDU (Jason Moore)
Subject: Re: SCI Out of Gas at Starting Gate?
Keywords: SCI
Organization: Oregon State University, Computer Science Department
Date: Thu, 30 Nov 1995 20:16:24 GMT
Message-ID: <49l3eo$e13@engr.orst.edu>

In article <49fk7s$mtc@sdcc12.ucsd.edu>, muller@sdcc33.ucsd.edu (Keith Muller) writes:

|> 
|> For the purposes of this analysis, all the assumptions err on the
|> conservative side.  
|> 

|> o  12 Million * .5 = 6M cache cycles/sec/quad PentiumPro on SCI
|>                ^^^ probability of cache line being on local node


OK.  I'll give it a try.  You make a very general assumption about locality.
Hopefully data are distributed reasonably enough that you don't need to access
remote memory half the time.  Of course, if you are assuming a naive data 
distribution, 50% may be right.

Jason


-- 
--------
Jason Moore                       Internet:  moorej@research.cs.orst.edu
Department of Computer Science    Bell-net:  (503) 737-4052
Oregon State University
"Anything worth doing is worth doing with a smile"


