Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Posting-Version: version B 2.10.2 9/17/84 chuqui version 1.7 9/23/84; site daisy.UUCP
Path: utzoo!watmath!clyde!cbosgd!ihnp4!mhuxn!mhuxj!mhuxr!ulysses!allegra!bellcore!decvax!ittvax!dcdwest!sdcsvax!sdcrdcf!hplabs!nsc!daisy!david
From: david@daisy.UUCP (David Schachter)
Newsgroups: net.micro,net.micro.pc
Subject: Re: Standard, What standard???
Message-ID: <77@daisy.UUCP>
Date: Wed, 27-Feb-85 22:49:19 EST
Article-I.D.: daisy.77
Posted: Wed Feb 27 22:49:19 1985
Date-Received: Mon, 4-Mar-85 04:58:03 EST
References: <143@idmi-cc.UUCP> <810@sjuvax.UUCP> <56@daisy.UUCP> <287@cmu-cs-k.ARPA>
Reply-To: david@daisy.UUCP (David Schachter)
Organization: Daisy Systems Corp., Mountain View, Ca
Lines: 28
Xref: watmath net.micro:9575 net.micro.pc:3430
Summary: 


Mr. Nowatzyk of Carnegie Mellon states that 64K segmentation limits have
caused him problems in using 80286 software on our workstations.  If he is
running very large designs through our older software, this can happen.
This has been corrected in newer releases in those places where it has
caused problems.  (When we designed the software, we designed it with what
we thought were generous safety margins.  Our customers promptly used the
increased efficiency of computer aided engineering to do much larger designs
than before!  Parkinson's law strikes again.)

All of the newer software, particularly in the physical layout tools and
in the hardware accelerator realm have taken advantage of what we learned
in doing the older software.  (That's what I meant in my earlier posting
when I  used the term "experience.")  We learned, in short, how to
design our code to run in the method intended by the designers of the CPU.
If you want to get maximum performance on a CPU you didn't design, this is
always a requirement, be it a NS32000, an MC68000, an 80286, or a PDP-8.

In our experience writing CAE software, in the rare cases where 64K segmentation
is a problem, it usually means that we don't know what we are doing yet.  There
is almost always a better algorithm that we haven't discovered yet, one which
uses smaller data structures >faster<.

Large address spaces are convenient.  They are not essential.  Moreover, their
convenience can rob you of the incentive to get maximum performance.  The
Intel architecture is a dark cloud with a silver lining: the need to keep within
the small address space frequently causes us to find solutions that are smaller
and faster, helping us meet our performance goals.