Re: [tablix-list] G-Tablix 0.0.8 release

From: Tomaž Šolc <tomaz.solcREMOVE@THISsiol.net>
Date: Thu Mar 24 2005 - 22:41:01 CET

Hi

> Just curious why you need to setup a PVM Cluster to run scheduling
> software? Are there any other applications you need PVM for?

Finding a solution to a timetabling problem can take a lot of processor
time. Even quite simple timetabling problems can take a lot of time to
solve on a single processor. Also Tablix is meant for schools that do
not have the latest in computing technology available to them. PVM
makes it possible to use a lot of older computers (like computer is the
computer science classroom) instead of a single powerful machine. The
genetic algorithm Tablix uses was quite simple to adapt to run
efficiently on a PVM cluster.

PVM is used by a lot of CPU intensive free software. For example Povray
(raytracer), Transcode (video transcoding tool), etc.

> I'm just now exploring the world of Linux CLusters for the first time
> and would like to introduce some examples to my students (AP Computer
> Science with java and linux). Eventually I'd like to teach them to
> write their own object oriented apps in java that take advantage of
> some sort of clustering paradigm.
>
> I know that JVMs are very hard to migrate. So, I'm looking into PVM
> or MPI with C for number crunching apps (ie: finding very large
> primes) or C and openGL for graphic apps (like rendering a highly
> detailed fractal). I'm also looking into openMOSIX with C++ and
> fork() methods (does openGL work with C++?).

I don't know much about Java. As far as I know, openmosix is a kernel
mode solution that does not need any special application support. You
only have to make your application multi-process or multi-threading.

> Last year we setup an openMOSIX Cluster (clusterKNOPPIX or QUANTIAN)
> on our LAN (25 Dell OptiPlexes with PIVs and 750MB RAM with
> fast-ETHERNET) and demoed existing clustered apps (KANDEL, POVRAY) but
> had a hard time writing our own clustered code....
>
> Any insight you could provide into these issues would be greatly
> appreciated!

I found the use of PVM quite straightforward. When I was developing the
first version of Tablix, I looked into PVM, MPI and a few other
clustering toolkits and I found PVM to be the easiest to use. I learned
how to use it in Tablix in about a week using the excellent
documentation and example code.

Best regards
Tomaz Solc
Received on Thu Mar 24 22:35:41 2005

This archive was generated by hypermail 2.1.8 : Tue Aug 16 2005 - 20:43:12 CEST