Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Path: utzoo!utgpu!water!watnot!watmath!clyde!rutgers!ames!ptsfa!ihnp4!alberta!sask!reid
From: reid@sask.UUCP
Newsgroups: comp.arch
Subject: The serial bottleneck
Message-ID: <648@sask.UUCP>
Date: Sun, 22-Mar-87 18:08:43 EST
Article-I.D.: sask.648
Posted: Sun Mar 22 18:08:43 1987
Date-Received: Wed, 25-Mar-87 01:19:39 EST
References: <492@cpocd2.UUCP> <43700014@uicsrd> <505@cpocd2.UUCP>
Organization: University of Saskatchewan
Lines: 25
Summary: There are inherently serial problems

In article <505@cpocd2.UUCP>, howard@cpocd2.UUCP (Howard A. Landman) writes:
> We have an existence proof that the serial bottleneck can be made to go
> away.  It's called the human brain.  So there's still hope!
> 	Howard A. Landman
> 	...!intelca!mipos3!cpocd2!howard


If you check out some of Hillis' Connection Machine references (like the
ones in the recent (Jan. 87, I think) CACM), he points out that there are
problems which are provably serial - that is, which cannot be speeded up no
matter how many processors you have.

The simple solution is just to not run these problems on your
super-multiprocessor.

Another recent poster made a more important point - when large-scale
multiprocessors become general-purpose enough to handle context switching, a
serial section in one process will simply free up all the other processors
to run something else at the same time.

 - irving reid -
-- 
reid@sask.uucp                          {alberta, ihnp4, utcsri}!sask!reid

What the world REALLY needs is a good Automatic Bicycle Sharpener.