Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Posting-Version: version B 2.10.1 6/24/83; site seismo.UUCP
Path: utzoo!watmath!clyde!floyd!vax135!cornell!uw-beaver!tektronix!hplabs!hao!seismo!mo
From: mo@seismo.UUCP (Mike O'Dell)
Newsgroups: net.arch
Subject: Stack caching
Message-ID: <727@seismo.UUCP>
Date: Wed, 21-Mar-84 09:34:15 EST
Article-I.D.: seismo.727
Posted: Wed Mar 21 09:34:15 1984
Date-Received: Sun, 25-Mar-84 11:49:42 EST
Organization: Center for Seismic Studies, Arlington, VA
Lines: 20

Yes indeedy, Burroughs did it LONG ago.  The B5500 was the first machine
marketed with Virtual Memory (although Burroughs was afraid noone would
understand that, so they didn't mention it) and it did no serious
caching as we understand it today.  The B6700 and B6800's did cache the
top two cells of the stack, I believe, while the B7800 maintained a cache
of the top 32 words or so in bipolar ram.  When the cache started to
overflow or underflow, it did block moves to or from memory.  Caching
a stack is very easy - you always know exactly what it can do.  I believe
the B7800 also had an instruction cache which worked in the more normal
LRU-ish kind of way, but between the instruction prefetch unit, the
normally small code segment size observed on the Burroughs, and the
large cache, the machine pretty much runs at cache speeds.

For more insight into machines clearly Ahead of Their Time, see
Elliot Organick's wonderful book "Computer System Organization:
The B5700/B6700"  Academic Press 1973.  This book should be required
reading for anyone writing documents describing a computer architecture.

	-Mike O'Dell