info-vax@ucbvax.ARPA (03/04/85)
From: Reed B. Powell <POWELL@DEC-MARLBORO>
We are investigating the methods useful to the market for connecting
the various "supercomputers" in the industry (Cybers, Crays, etc) to
Digital's clustered systems. There are many reasons people have for
desiring this, but usually it falls into the general catagory of
"front-ending" the expensive supercomputers with lower cost
systems used for data-preperation, data-presentation, etc. IE: Keep the
supercomputer busy doing what it is best at.
There are a number method of doing this. Hyperchannel seems to be doing
well in the most general case of connecting anything to anything, but
also seems to be adversely affected by coping with those generalities.
What DEC is looking at is a method of hooking those systems up to
DEC's systems.
The only thing we have pretty much decided -- and even this is not cast
in concrete -- is that is should be oriented around a connection to the
CI bus, which is what connectes the systems of a cluster, as well as
their disk servers, together.
What we are specifically soliciting is information in two broad areas:
1. What functionality does the market want? Should we look into
servers, gateways, rje-s, or what? Should the supercomputers
"look" like another DEC computer on a cluster, or should there
be something in between that performs a "monkey in the middle"
function.
2. Ideas on implementation. It would be best if anyone submitting
ideas in this catagory also include some information from the
above catagory, since we must in the end justify our actions
based on market needs.
Since most of the supercomputers are in sites related to ARPAnet,
this seems to be the best, most effecient way to gather this data.
Anyone wanting to talk directly about this should leave a phone number,
and someone related to this project will be glad to talk.
thanks in advance,
-reed
--------info-vax@ucbvax.ARPA (03/05/85)
From: Ron Natalie <ron@BRL-TGR.ARPA> We are in the process of acquiring a (yet to be determined) supercomputer to integrate into our collection of mostly UNIX machines. The machine must come with an appropriate network hardware to interface the super computer to three UNIBUS's, two SELBUS's, and an IBM channel. Looks like Hyperchannel is the one people are going to be bidding. Personally, I detest NETIX. In order to simplify software on the NON-Supercomputer operating systems, and to facilitate communications between these six core minis and the rest of the BRL complex, we have specified that the protocol must be TCP/IP. We already use it extensibly, and have a public domain spooling and rje system called MDQS that we rely on heavily for our internal use (we only have three or four lineprinters and two laser printers to split among a whole ton of machines). Supporting DEC Cluster is a nice idea, looking at the DEC<->Super problem narrowly, but I think that a more generic DEC CLUSTER<->WORLD interface would be more useful. Especially with people with such diverse supercomputers as ours (we already have two from different manufacturers and it looks like the next one will go to even a third). -Ron
info-vax@ucbvax.ARPA (03/05/85)
From: Mike Muuss <mike@BRL-TGR.ARPA>
BRL is currently engaged in detailed planning for connection
of various computers to a > 100 MIPS supercomputer (Cray-2, ETA 10, etc).
Our strategy uses TCP/IP based networking and standard + local
applications protocols to tie it all together.
TCP and IP have the advantage that they work over any communications
hardware, and are MIL Standards (MILSTD 1777, MILSTD 1778).
Best,
-Mike Muuss
(301)-278-6678
AV 283-6678
FTS 939-6678
ArpaNet: Mike @ BRL
UUCP: ...!{decvax,cbosgd}!brl-bmd!mike
Postal:
Mike Muuss
Leader, Advanced Computer Systems Team
Computer Science and Mathematics Branch
Systems Engineering and Concepts Analysis Division
U.S. Army Ballistic Research Laboratory
Attn: AMXBR-SECAD (Muuss)
APG, MD 21005info-vax@ucbvax.ARPA (03/20/85)
From: SAMUELSON%LLL@LLL-MFE.ARPA
From: Norm Samuelson <Samuelson@LLL.MFENET>
To: Powell@DEC-MARLBORO.ARPA,
tops-20@SU-SCORE.ARPA,
info-vax@SRI-KL.ARPA
In-Reply-To: Message from "Reed B. Powell <POWELL at DEC-MARLBORO>" of Mon 4 Mar 85 12:37:00-PST
Telephone: (415) 422-0661, [FTS 532-0661]
Office-Location: Bldg 543, room 1067
Postal-Address: LLNL, PO Box 5511, L-630, Livermore, CA 94550
Since, as you noted, most CRAY sites use HYPERCHANNEL to connect
their CRAYs to each other, you should come up with an interface
from the CI to the HYPERCHANNEL, not try to replace the Hyperchannel.
- Sam -
-------