Newsgroups: comp.parallel.pvm
Path: ukc!uknet!comlab.ox.ac.uk!comlab.ox.ac.uk!nasng
From: nasng@comlab.ox.ac.uk (G N Shrinivas)
Subject: PVM 3.3
Message-ID: <1994May31.101727.3429@nagp19.comlab.ox.ac.uk>
Originator: nasng@nagp19.comlab
Keywords: pvm3.3 sharedmemory
Sender: nasng@comlab.ox.ac.uk (Shrinivas Gorur)
Organization: Oxford University Computing Laboratory, UK
Date: Tue, 31 May 1994 10:17:27 GMT
Lines: 30

Hi,

Is pvm3.3 out yet?? Does anybody have any ideas on when it should be out. I
had heard that it should be out sometime soon.

Just for information...

I have used pvm3.2.8(sharedmem) version on a SGI challenge and found it to
have a very erratic behaviour in the time it took to send and receive
messages. Also it wasn't very much quicker as I thought it might be.

An additional problem I observed was that when I used the 3.2.8 and 3.2.6
together (to form a PVM) and used the groups feature, it used to hang at
times and the pvmgs used to go into some sort of a loop and I had to kill
it and restart it. Similar problems weren't observed when pvm3.2.6 was used
on the sharedmem machine. So i guess it was a pvm3.2.8 problem or a
3.2.6_talking_to_3.2.8 problem.

As a comment, I hope more hardware manufacturers come out with their own
implementations of pvm. To expect a very generic off the shelf software to
have excellent performance one each architecture is rather ludicrous. PVM has
certainly opened new vistas but now the hardware manufacturers should provide
pvm with best performance on their machines. Or as somebody suggested, a 
commercial release which does the above should be pursued by the PVM team 
(come on you can rake in your $$$$$s).

Regards,
shrinivas



