Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Posting-Version: version B 2.10.2 9/5/84; site alberta.UUCP
Path: utzoo!utcsri!ubc-vision!alberta!jim
From: jim@alberta.UUCP (Jim Easton)
Newsgroups: net.sources.d
Subject: Re: P & V Algorithms needed
Message-ID: <857@alberta.UUCP>
Date: Fri, 21-Mar-86 16:59:33 EST
Article-I.D.: alberta.857
Posted: Fri Mar 21 16:59:33 1986
Date-Received: Sat, 22-Mar-86 08:06:33 EST
Distribution: net
Organization: U. of Alberta, Edmonton, AB
Lines: 28

***

>I have never heard of a mutual exclusion scheme that did not
>at root depend on the existence of an interlocked test-and-set
>operation of some kind.  I would be interested in hearing some
>details about other schemes if they exist.

It is a sad commentary to note that someone has to go and "invent" a
special instruction for this when most of the old machines had one that
was ideally suited for the purpose and was useful for other things as well.

The instruction was "exchange memory and A (register).  The only
requirement from the hardware is that it cannot be interrupted in
mid execution - unavoidable with core memory.

I believe that key to this problem is to have only one "item" that cannot
be duplicated.  Every process makes a grab for it but the one that gets
it is the one that gets the resource.

I personally call that "item" a "ticket".  It is like a hockey ticket - the
person who has it gets to sit in the corresponding seat at the game and the
people that issue the tickets know that there cannot be collisions because
there is only one ticket (/seat/game).

Anyway that's how I think of it and if the machine has a way of arranging for
a "ticket" the algorithm will work - otherwise I wouldn't bet on it.

	Jim Easton (..!alberta!jim)