Path: utzoo!mnetor!uunet!husc6!bloom-beacon!tut.cis.ohio-state.edu!mailrus!ames!pasteur!ucbvax!richter.UUCP!krowitz
From: krowitz@richter.UUCP (David Krowitz)
Newsgroups: comp.sys.apollo
Subject: Re:  Giant operating systems
Message-ID: <8803302015.AA22036@EDDIE.MIT.EDU>
Date: 30 Mar 88 18:50:48 GMT
Sender: daemon@ucbvax.BERKELEY.EDU
Organization: The Internet
Lines: 30

Here, here!!! I agree! I can remember working on a DEC-10 mainframe
that only had 64K words (about 256k bytes) of memory, total. I even
wrote a real-time, multi-tasking OS for a PDP-8 that fit in less than
8K words. This all requires some carefull planning and good, old 
fashioned programming by people who understand the hardware (something
which is getting increasing rare these days) -- things like grouping
subroutines which call each other onto the same memory pages so to
reduce the VM paging for the OS, actually counting the number of
assembler instructions produced by the compilers, and analyzing the
assembly code to see if it can be shrunk in size/speed up. Most
programmers get taught Lisp, and Pascal in school, learn C out of
necessity, and have never seen the assembler for the machine they
are working on. Apollo doesn't even have a hardware architecture 
manual for their machines, and the assembler (they do have one) is
not a realeased product. The philosophy of the day is that programming
is to provide features, and hardware is to  provide performance.
Given the relative costs of engineers/programmers and a new DN3000,
that is not a bad philosophy (I cost my boss over $60,000 a year
when you through in MIT's overhead and benefits, and DN3000 can
be had for $3500); but *skilled* programming can provide *both*!


 -- David Krowitz

krowitz@richter.mit.edu   (18.83.0.109)
krowitz%richter@eddie.mit.edu
mit-erl!mit-richter!krowitz@eddie.mit.edu
mit-erl!mit-richter!krowitz@mit-eddie.arpa
krowitz@mit-mc.arpa
(in order of decreasing preference)