Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP Posting-Version: version B 2.10.2 9/18/84; site utcsri.UUCP Path: utzoo!utcsri!Jack From: Jack Newsgroups: ut.supercomputer Subject: In response to Alan Karp's comments Message-ID: <2110@utcsri.UUCP> Date: Wed, 12-Feb-86 22:31:26 EST Article-I.D.: utcsri.2110 Posted: Wed Feb 12 22:31:26 1986 Date-Received: Thu, 13-Feb-86 01:27:35 EST Sender: daemon@utcsri.UUCP Organization: CSRI, University of Toronto Lines: 32 Resent-Message-Id: <8602130329.AA16105@utcsri.UUCP> To: supercomputer%nyu.arpa@CSNET-RELAY Resent-From: Lou Salkind Resent-To: supercomputer%nyu.arpa@CSNET-RELAY Resent-Date: Wed, 12 Feb 86 11:46:47 est In Alan Karp's summary of the recent ARO conference at Stanford on medium scale parallel computers he left the impression that "there is currently a dearth of commercially available, shared memory systems. " We would like to offer a different impression of the availability and usefulness of shared memory parallel computers that one can purchase commercially today. In addition to the Alliant, Flex32, and Elxsi systems that were mentioned there are a number of other very successful shared memory machines including BBN Butterfly (256 processors) Sequent Balance 16000 (24 processors) Encore Multimax (20 processors) CRAY X-MP-4 (4 processors) CRAY 2 (4 processors) We also must remember that while the Denelcor HEP was a commercial failure, it certainly was a success in terms of providing a parallel computer that was programmable and capable of doing real computations. We have had experience with the Alliant, Sequent, Encore, CRAY X-MP and Denelcor machines. All of these computers out perform existing cubes and in our view are easier to program and debug. If you want to account for cost/performance place Sequent and Encore against the Intel's iPSC and Alliant against Ncube. Moreover, these systems (Alliant, Sequent, Encore) are stand alone minicomputers that can perform the ordinary computing one would normally does on say a VAX system including sharing resources among a number of users. For these reasons we fail to see any dominance of the cube architecture at this moment. >From the standpoint of developing algorithms, the hypercube does provide an interesting architecture to investigate. Jack Dongarra Dan Sorensen