Scalapack scalability and lu back substitution info
Posted: Mon May 26, 2008 6:58 amHi ppl
I am using scalapack to solve a linear system the thing is that i need the system to be solved multiple times.
The system is Ax=B.The matrix A remains unchange while matrix B changes over itteration.
I want to solve the system multiple times for different B's.In the algorithm that i use
1)I decompose the matrix A using scalapack
2) start the itterations and then doing back substitution for different B's
I am using scalapack to make the back substiturion and it seems to me that the back substitution takes longer than expected. Any ideas on the grid that i should use,Should it be square?And what about the block size.I think that the computational load should be low for back substitution so the messages should be as large as possible.
Any ideas would be i appreciated
I am using scalapack to solve a linear system the thing is that i need the system to be solved multiple times.
The system is Ax=B.The matrix A remains unchange while matrix B changes over itteration.
I want to solve the system multiple times for different B's.In the algorithm that i use
1)I decompose the matrix A using scalapack
2) start the itterations and then doing back substitution for different B's
I am using scalapack to make the back substiturion and it seems to me that the back substitution takes longer than expected. Any ideas on the grid that i should use,Should it be square?And what about the block size.I think that the computational load should be low for back substitution so the messages should be as large as possible.
Any ideas would be i appreciated