Newsgroups: comp.parallel.pvm
From: subraman@cs.iastate.edu (Krishnan Subramaniam)
Subject: Re bandwidth measurements on Ethernet...
Organization: Iowa State University, Ames, Iowa
Date: 27 Mar 95 05:44:05 GMT
Message-ID: <subraman.796283045@shazam.cs.iastate.edu>

Hello,
	I've some questions on how and what do we consider bandwidth
measurement on a parallel system. 

1. Do we measure bandwidth of the network alone by measuring time for 
a send/recv like in the timing program that comes with PVM ??

2. Should we not consider the effective bandwidth when we intersperse
computation and communication ? Because almost no parallel program
is going to be merely a data transfer program, and so there will be
some mismatch in when the sender initiated the send and when the reciever
initiated the recv - will this affect bandwidth seen by the app.

3. If we know T_comp for the code in a parallel prog, then is it
legal to just say that T_comm = Total time - T_comp and use this
to compute bandwidth ?

4. Is it fair to use the numbers from programs like the timing program
as the bandwidth used effectively by PVM ? It is merely transferring
data - more as an ftp would - so is it still valid to say that using
PVM we can achieve this % of the peak bandwidth ??

	Some of the questions above may seem repetitive/redundant but
I just wantd to be (hopefully) clear. 
Hope I get some interesting replies ...

Thank you all,
Krishnan.

--
Krishnan R Subramaniam
Voice Mail : 515 - 296 4209.
Email	   : subraman@cs.iastate.edu   
HTML  	   : http://www.public.iastate.edu/~rsk/

