Newsgroups: comp.parallel.pvm
From: Aaron Jackson <shadow@apricot.com>
Subject: Changing PVM
Organization: TZ-Link, a public-access online community in Nyack, NY.
Date: Fri, 17 Nov 1995 17:25:05 -0600
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
Message-ID: <30AD19D1.FF6D5DF@apricot.com>

Hey there,

I'm in the process of making a number of changes to PVM because I'm
using it as a backbone for something it was not built for.  But in
doing all of this, I think there are some interesting points that
may come up for future people to pull from.

1) What happens if more than one person of different uid would
   like to talk to a single pvmd?  I have changed much of the
   pvm/pvmd guts to use instances instead of uid... not elegant
   but I'm in rapid mode so it works.  Unfortunately, it means
   that the t-d auth protocol goes right out the window.

2) We really only need to validate task-pvmd if it is accessing
   some important piece of information.

3) Task-task protocol is TCP in some cases.  This can be rather
   slow and I suspect a nasty bottleneck for pvm_mcast(), I now
   use RMP which give me multicast and authentication (bonus!).

Does anyone else see what thing might want to be done.  FYI, I'm
building a scaleable data distribution backbone on top of PVM.
And while its not of the nature of most PVM problems, this nails
many problems in traditional distributed systems on the head.

Comments requested, please dont send mail, the machine's gonna
be offline for a little while.

-- Aaron

