Newsgroups: comp.parallel.pvm
From: Stuart D Blackburn <sdblackb>
Subject: Re: Maximum Buffer Size?
Organization: University of NC at Charlotte
Date: 21 Aug 1995 17:39:17 GMT
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
Message-ID: <41agc5$dts@news.uncc.edu>

I have had some experiences with PvmDataInPlace which seems to
indicate that when using this method of buffering that PVM sends a separate
network packet for each piece of data that you pack. For example, if you 
are sending an two integers (named num1 and num2) and an array of floats
(named foo). You use 3 separate pack instructions to pack up this message,
and it seems that these 3 pieces of data end up packed into separate network
packets for delivery to the destination, where they are combined into the
message the PVM expects.
	I point this out because if this is way that PVM is sending messages with
PvmDataInPlace, then there may indeed be a limit to the size of each
piece of data that you pack. I would definately try to use one of the other
buffering schemes, probably PvmDataRaw, to see if this causes the same problem.
	I have passed 50x50 arrays of floats using PvmDataRaw, so I think
this should work. If not, you should carefully check to make sure that you
are sending and receiving the same number of elements particularily if you
are using pvm routines like pvm_scatter and pvm_gather.

	Good Luck,
		Stuart Blackburn


