From: gw334@cleveland.Freenet.Edu (Konstantyn Kharchenko)
Newsgroups: comp.parallel.pvm
Subject: solving sparce linear system on the network?
Date: 15 Oct 1998 06:23:21 GMT
Organization: Case Western Reserve University, Cleveland, OH (USA)
Message-Id: <7044cp$nfl$1@pale-rider.INS.CWRU.Edu>
Reply-To: gw334@cleveland.Freenet.Edu (Konstantyn Kharchenko)


Dear PVM users,

I would like to discuss  some questions about solving sparce
system of linear equitions using the PVM.

The main  problem  is it  possible to solve  a sparce linear
system on the network cluster (uniprocessor computers on the
net) faster then a sequential version?

I had many chances to try  PETs and  ScaLAPACK  on 10Mb  and
100Mb Ethernet and couple of Pentium computers but I  didn't
reach  better  result  then sequential  version. (Of  course
I  understand I can make something wrong to run it in proper
way).

What  do  you  think  about  teoretic chance to solve linear
system faster on the 10Mb Ethernet cluster with  the latency
about 1ms?

If you have  any experience how to reach good performance on
solving linear equitions by PVM on  10-100Mb Ethernet please
just tell it is possible...

Thanks in advance for your suggestions.

Sincerely,
Konstantyn

konst@cad.ntu-kpi.kiev.ua