Newsgroups: comp.parallel.pvm
From: bwh@inf.rl.ac.uk (Brian Henderson)
Subject: Re: PVM under Windows?
Organization: The Unconfigured xvnews people
Date: 25 Aug 1994 12:24:08 GMT
Message-ID: <33i2h8$2shs@unixfe.rl.ac.uk>


Cripsin Cowan incorrectly summaried my previous posting as:
... >bwh@inf.rl.ac.uk (Brian Henderson) writes:
... >: [PVM on Windows is useless because Windows doesn't do pre-emptive
... multitasking]

This not a true reflection of what i said, as i dont believe PVM on windows
to be "useless" but to be "less practical" within an office/department environment
where you are seeking to use the "idle-cycles" of other peoples machines.

In article shj@cantaloupe.srv.cs.cmu.edu, pdinda@cs.cmu.edu (Peter A Dinda) writes:
>  A reasonable requirement could be that PVM programs be compiled into
>  DPMI DOS applications or possibly Windows VDDs, both of which are
>  pre-emptively round-robbined by Windows. 

I was unaware that the DPMMI/VDD device driver route can give some form of
pre-emptiveness but is this GUARENTEED to run in any given time period or
when the current process manually frees itself?

> Another possiblity is to "instrument"
>  the most frequently used PVM functions with liberal calls to yield() and
>  compile PVM programs as windows programs.  The first option is probably
>  the preferable one

As to "yield()"-ing: most PVM progs i have seems have the form

		pvm_recv()	: get data

		process_it()	: for longgggggggggg time

		pvm_send()	: send answer

its during the "process_it" stage when the whole PC hangs with no, screen update,
user input, clock update........ Yielding doesn't help the calculation stage.

>, but porting PVM to Windows does appear possible.
>
>

 Possible, yet no one yet done it? If not why not? Theres a world-wide market
ready for such a solution yet no takers....mmmm

 Were just beginning to look at the possibility of PVM on NT and this has severe
problems itself -let alone looking at WINDOWS.

I would like to see PVM on WINDOWS/WINDOWS-NT as then codes code port across
large super-computers, MMP systems, workstation clusters and PC clusters.

---
Brian W. Henderson		    *	UK JANET: bwh@uk.ac.rl.inf
Parallel Processing Group	    *	UUCP:	  ..!mcsun!ukc!rlinf!bwh
Computational Modelling Divison,    *
Informatics Department,		    *******************************************
Rutherford Appleton Laboratory,
Chilton, Didcot, Oxon. OX11 0QX,	Tel : +44 + (0235) 821900 ext. 6151
_______________________________________________________________________________



