Newsgroups: comp.parallel.pvm
From: papadopo@cs.utk.edu (Philip Papadopoulos)
Subject: Re: Triangle Inequality?
Organization: CS Department, University of Tennessee, Knoxville
Date: 31 Aug 1994 09:01:57 -0400
Message-ID: <341v05INNh5p@duncan.cs.utk.edu>

In article <33vj5e$2m1@falcon.ccs.uwo.ca> crispin@csd.uwo.ca (Crispin Cowan) writes:
>PVM guarantees in-order delivery of messages between pairs of tasks,
>but does it guarantee the triangle inequality?  I.e. if we have three
>tasks A, B, and C.  A first sends a message directly to C, and then A
>sends a message to B which in turn forwards it to C.  Is it guaranteed
>that the direct message will arrive before the indirect message?
No way.  One never knows what network delays will be encountered when
sending a message.  PVM guarantees order by putting a serial number
on each message. That way if a message arrives at a destination process
with an incorrect serial number, the receiver will wait until the 
correct message is sent.  With the intermediate stop at B, there is
no way tell whether the message from A or the message from B will
arrive first at C.


-Phil Papadopoulos


