[Trilinos-Users] solving N linear systems simultaneously on parallel platform

[Heroux, Michael A]

Rama,

In order to solve each individual system you do not need an Epetra_MpiComm object.  You would start up your parallel run using MPI and each MPI process would create an independent Epetra_SerialComm object, use it to create any maps that are needed, usually just the rowmap, use the map(s) to create vector and matrix objects, preconditioners, etc. and then solve each independent problem without any need for communication.

If your intent is to combine results from the individual solutions in parallel, you will need something more complicated, but getting the above mentioned code working is a good first step.

Mike


On 10/10/08 1:49 PM, "RAMAKRISHNA TIPIREDDY" <tipiredd at usc.edu> wrote:

Hi Mike,

Thank you for your quick response. Could you please elaborate on this. I was trying to use, serial comm and have some confusion.

Do I need to use both mpi_comm and serial_comm or just serial comm for each processor.

My objective is to use the solutions from all linear systems to modify the rhs of each linear system for next iteration.

Thanks & Regards,
Ramakrishna Tipireddy.



On 10/9/08, Heroux, Michael A <maherou at sandia.gov> wrote:
Rama,

Since you want each processor to work independently, you don't need to distribute your matrix. Instead on each processor create a serial comm, related map, etc.

Is this clear?

Mike

________________________________
From: trilinos-users-bounces at software.sandia.gov
To: trilinos-users at software.sandia.gov
Sent: Thu Oct 09 12:04:46 2008
Subject: [Trilinos-Users] solving N linear systems simultaneously on parallel platform

Hi all,

I am trying to solve N linear systems simultaneously,

K_i u_i = f_i,  for i = 1 to N,

I would like to solve this problem on parallel platform such that each processor solves one linear system. ie ith processor solves K_i u_i = f_i.

I used Epetra_CRSMatrices to distribute rows of a matrix on to different processors. I am just wondering how to distribute each linear system on to diiferent processors.

Thanks in advance,
Ramakrishna Tipireddy



-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://software.sandia.gov/mailman/private/trilinos-users/attachments/20081013/b4390ef7/attachment-0001.html