kar@ritcv.UUCP (Kenneth A. Reek) (05/17/84)
Subject: Re: Pyramid comes through for HHB (No more restraints !!)
<>
In answer to some recent comments I've received: Yes, the imple-
mentation of a machine, while not dictated by the interface, is certainly
driven by the needs of the interface.
The discussion seems to have become somewhat moot now that Pyramid
is supporting non-aligned longword accesses. This was accomplished,
as reported by Bob Erickson of HHB-Softron, by a change to the microcode.
The performance penalty paid is at most 5%, probably none at all. Their
rational for making the change seems to be the marketplace -- Bob tells us
that they figured their machine would be able to run more existing software
this way. This backs up my original argument that needless architectural
differences between machines should be eliminated because they hinder
portability.
Hmmmm.... Microprogramming appears to be a much more flexible
method of implementation than a lot of us had previously thought....
(Myself included -- I was surprised it could be done that easily.)
Ken Reek, Rochester Institute of Technology
{allegra,seismo}!rochester!ritcv!karmark@umcp-cs.UUCP (05/18/84)
Now if only Pyramid would offer a microcode option to address bytes
in DEC order instead of IBM order, they would really have a killer
machine.
--
Spoken: Mark Weiser ARPA: mark@maryland
CSNet: mark@umcp-cs UUCP: {seismo,allegra}!umcp-cs!markbob@hhb.UUCP (05/19/84)
<< "All in one moment, and so near the brink" - Milton, Paradise Lost >>
Note: This news is far more relevant if you have been following the
dialogue about "Pyramid architectural restraints" started by
myself on April 22.
After discussing the problems HHB-Softron was having with getting the
CADAT simulation system to run on the Pyramid 90x due to the longword
alignment problem, Pyramid has agreed to change the microcode of the
machine to allow for longword moves from even, non-quad byte
boundaries. (.e.g 2,6,10...). Pyramid also feels that by making this
change, they will be allowing other major applications which might
exhibit this same feature to run on their machine.
Pyramid assures us that this will have a minimal impact on the overall
performance of the machine. (Guaranteed no greater that 5%.) And they
believe it will have have no impact at all.
We are now the proud owners of a Pyramid 90x system with 8MB of memory,
Data Cache, 64 ports, Ethernet and the dual port 4.2/Sys5 operating
system. Thanx to the responsiveness Pyramid has shown us, the CADAT
concurrent fault simulator will run like the wind on one of the best
price/performance UN*X machines now available.
Those of you who argued that to have a machine which supported this
feature required lots of expensive hardware ought to reconsider. Isn't
it amazing what a little well designed harware and microcode can do ?
The Pyramid 90x is definitely one less machine we have any right to
ignore...
========================================================== Be
Bob Erickson
Company: HHB-Softron
UUCP address: {decvax,ihnp4,allegra}!philabs!hhb!bob