From: Anders Jorgensen <ajorg@lanl.gov>
Newsgroups: comp.parallel.pvm
Subject: Scheduler
Date: Mon, 15 Mar 1999 12:30:31 -0700
Organization: Los Alamos National Laboratory
Message-Id: <36ED5FD7.64D48E59@lanl.gov>
Reply-To: ajorg@lanl.gov
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
Xref: ukc comp.parallel.pvm:8120


Dear  PVMers,
	I am in need of a PVM resource manager that can do fault-tolerant
scheduling. The scheduler will receive a PVM packet from the main
program, and will be responsible for the processing of the data and
return of results to the main program.

1) The scheduler keeps a list of all packets that have not yet been
processed. Or perhaps optionally a "receipt" so that it can go and
re-request the packet from the main program should it need it.

2) When a host goes down, packets that were being processed on that host
must be resubmitted to other hosts.

3) When the main program sends a certain message (a block) to the
scheduler, the scheduler must get all packets processed as quickly as
possible, by resubmitting not-yet-completed packets to hosts as they
free up.

4) Scheduler must also on its own try to re-attach hosts that drop off
the network, so that during a long computation the network doesn't keep
shrinking.

5) It would be nice if the scheduler could add and delete hosts based on
time of day (to remove certain peoples machines from the VM during
working hours).

I am thinking of writing this piece of code for a project I'm working
on, but I thought I would check first whether anyone else has written
something that does some or most of the functions I mentioned.
                           ----anders