Re: ||ism

[Kevin Crowston]

>1.  I think we need a || methodology that will support *both*
>distributed computing, where synchronization may be seldom required,
>and || computing, where synchronization may occur very often.  This
>seems to imply to me that message passing will become a serious
>hindrance in the latter.  And that makes me think that we might not
>want to dive glibly into using a system based on MPI.

I agree that we want to support both approaches, but I don't think it
follows that MPI is necessarily the wrong solution.

My impression of the reason for considering MPI was portability, not
because message passing was always the right approach.  My concern about
using shared memory is that it might limit the range of machines that could
be used.  For example, I don't believe that the IBM SP2 supports shared
memory (to take an example of a machine to which I have access).

MPI does claim to support both message passing and massively parallel
machines.  I believe that the intent is that it take advantage of shared
memory where available.  Therefore, massively parallel programs might okay
on a machine with shared memory, even if you build on MPI (and still run,
but slowly, on the SP2).  However, I have no idea how efficient any of
these implementations are.

This isn't to say that MPI is the right way to go; only that it solves one
of the problems facing Swarm, namely portability.  PVM is another
supposedly portable package, and there are the usual religious wars about
which is superior.  I believe PVM is also message passing though, since you
can implement message passing on a shared memory machine but not vice versa
(or at least, not easily).

Kevin




                  ==================================
   Swarm-Modelling is for discussion of Simulation and Modelling techniques
   esp. using Swarm.  For list administration needs (esp. [un]subscribing),
   please send a message to <address@hidden> with "help" in the
   body of the message.
                  ==================================