Path: utzoo!utgpu!news-server.csri.toronto.edu!cs.utexas.edu!wuarchive!zaphod.mps.ohio-state.edu!sol.ctr.columbia.edu!emory!hubcap!fpst
From: berryman-harry@CS.YALE.EDU (Harry Scott Berryman)
Newsgroups: comp.parallel
Subject: Re: New i860 parallel machine
Summary: i860 is too fast(or transputer is too slow)
Keywords: i860 parallel
Message-ID: <1991Apr22.122940.1907@hubcap.clemson.edu>
Date: 20 Apr 91 22:51:17 GMT
References: <1991Apr18.120233.339@hubcap.clemson.edu>
Sender: news@CS.YALE.EDU
Organization: Yale University Computer Science Dept., New Haven, CT 06520-2158
Lines: 41
Approved: parallel@hubcap.clemson.edu
Originator: berryman@perch.CS.Yale.Edu
Nntp-Posting-Host: perch.zoo2.cs.yale.edu


WARNINIG! I AM NOT A HARDWARE JOCK! TAKE MY HARDWARE COMMENTS WITH
                      A GRAIN OF SALT! 


In article <1991Apr18.120233.339@hubcap.clemson.edu> zenith@ensmp.fr (Steven Ericsson Zenith) writes:
>
>They've put together a desktop supercomputer out of a standard 386 PC,
>MSDOS, added UNIX System V, a bunch of transputers each of which connect
>and have shared memory access to a 40MHz i860. The company was founded
>by one of the guys who did the core work on Helios - so I'd expect them
>to live up to their "seamless" claim.

There are two serious problems I see with this kind of design. The first
problem is that the i860 will be far faster than the transputers which link
them. The second is that the i860 will be need heaps of fast memory or it will
be cache-bound. These two problems actually reer there ugly heads on the
iPSC/860. The network is not nearly as fast as the processors and the i860s
spend a lot of time waiting for memory. The two problems agrivate each other.
It is an intersting set of parts they've picked.

Custom CMOS may help us out of a bind on some of the parts. You can 
build a custom chip for under $100,000. For what target audience would we
be building such a beast for? A dbase machine has far different constraints
from a scientific machine. There is no Cray version of Ingress, last I heard.

As the whole world (at least my part of it) seems intent on building the
better scientific parallel machine, I suggest we talk about a different
set of problems. Two come to mind:

1) Reservation systems (like for airlines and hotel chains)
2) Digital signal processing (in real time)

I know very little about either application. Anyone out know any of the
requirements for such systems, or have suggestions for other problems?

I agree with you, Steve, this party's been a bit of bore lately. Let's
get some action out here.

Harry Scott Berryman 
Yale Computer Science Department and ICASE/NASA Langley Research Center
berryman@cs.yale.edu

-- 
=========================== MODERATOR ==============================
Steve Stevenson                            {steve,fpst}@hubcap.clemson.edu
Department of Computer Science,            comp.parallel
Clemson University, Clemson, SC 29634-1906 (803)656-5880.mabell