Path: utzoo!utgpu!jarvis.csri.toronto.edu!mailrus!tut.cis.ohio-state.edu!ucbvax!decwrl!purdue!mentor.cc.purdue.edu!mace.cc.purdue.edu!mhm
From: mhm@mace.cc.purdue.edu (Michael Marsh)
Newsgroups: comp.sys.sequent
Subject: Re:Questions about Pyramid/Sequent
Keywords: multiprocessors,uniprocessors
Message-ID: <2136@mace.cc.purdue.edu>
Date: 5 Apr 89 20:55:45 GMT
Distribution: na
Organization: Purdue University
Lines: 28


	In comparing uniprocessors with multiprocessors, some years ago IBM
did a study that showed that there are situations where a multiprocessor
system has certain advantages over a monoprocessor that is "twice as fast".
These are in the areas of interrupts and context switching.  Large, fast
machines tend to get a lot of their performance from techniques such as
sophisticated caches, I/O processors, pipelining. large register sets, and
the like.  These mechanisms work most efficiently on large jobs using
sophisticated compilers which take advantage of the architectural oddities of
the particular machine. (Sound like the place to do batch processing? Yup!)
Context switching and interrupt processing break these machines, and that is
just what you get a lot of with 300 logged-on users.  As you add more users to
a large monoprocessor, a point will be reached where the system is spending
all it's time context switching.  A multiprocessor can handle more of this
activity before degrading (if designed properly).  A second advantage to the
multiprocessor architecture is that an interrupt need "interrupt" only one
processor, and in some implementations, (sequent's, for one) the interrupt
is handled by the processor running the lowest priority process, leaving the
more important processes to run uninterrupted.  On a monoprocessor, even high
priority jobs are prempted by interrupts.
	There is a psychological advantage to the multiprocessor in that the
performance tends to be very uniform over a wide range of loads, whereas a
uniprocessor, although faster, will start to slow down quicker, leaving the
users with the feeling that they aren't getting all they could out of the
machine.

	Disclaimer: These are the opinions of the author and not necessarily
those of Purdue University or Sequent.