Path: utzoo!attcan!uunet!tut.cis.ohio-state.edu!zaphod.mps.ohio-state.edu!van-bc!ubc-cs!uw-beaver!cornell!ken
From: ken@gvax.cs.cornell.edu (Ken Birman)
Newsgroups: comp.sys.isis
Subject: ISIS "homework" problem 2 (user-contributed)
Message-ID: <37283@cornell.UUCP>
Date: 13 Feb 90 15:29:34 GMT
Sender: nobody@cornell.UUCP
Reply-To: ken@cs.cornell.edu (Ken Birman)
Distribution: comp
Organization: Cornell Univ. CS Dept, Ithaca NY
Lines: 44

>From bernhold@qtp.ufl.edu (Dave Bernhold) Mon Feb 12 19:58:58 1990

Our situation is that we have a network of 50 Sun workstations with three
file servers and one compute server.  My estimate is that probably 50% or
more of the workstation cycles are wasted (even graduate students have to sleep
sometime :-).

Our work (quantum chemistry and physics) involves a lot of simple linear
algebra manipulations.  What I would like to see is a "linear algebra
server" using ISIS -- implement distributed algorithms for the basic BLAS,
LINPACK, EISPACK (soon LAPACK) routines and make a special set of libraries
which look to the caller like the usual BLAS etc. routines, but inside
they operate through ISIS to distribute the computation.

To avoid inflicting this on users whose workstations are only idle briefly,
I would set it up so that certain load average and inactivity criteria are
satisfied before a node can offer to be part of the "linear algebra server".

You are probably more familiar with it than I, but the Condor system from
Wisconson uses this kind of criteria to decide if it can accept a job or
not.  If you're not familiar, you can ftp it from shorty.cs.wisconsin.edu.

If you want another application idea from the world at large, a cross between
Condor and NQS (Network Queuing System) would be useful -- to provide a *real*
network queuing system, with network-wide queues, served by many hosts, etc.
This is the kind of thing I asked about a long time ago on comp.sys.isis with
regard to MDQS instead of NQS (they want $$ for NQS).

Unfortunately, I've not had time to work on either of these ideas...
I don't even have enough time to do my own work -- writing the programs
that would *use* the linear algebra server and run on the distributed
queuing system.  Sigh.  

Anyway, if anyone comes up with any applications along these lines, please
be sure to post about them in comp.isis.  And if I ever get time to work on
this stuff, you'll be among the first to know.

Dave Bernhold

[KB: I think this sounds like an excellant problem for someone who wants ]
[to get a little experience using ISIS without taking on something really]
[hard.  Although a reallyh fancy system along these lines would be hard  ]
[to design and build, something modest should really be pretty trivial.  ]
[Using META, it would be easy to identify unloaded/idle workstations...  ]