Peter Jay Salzman
p at belial.ucdavis.edu
Fri Jul 28 12:47:21 PDT 2000
i'm just beginning to learn about MPI programming.
i have a series of linear equations which are represented by a tridiagonal
matrix. the order of the matrix is between 1000 and 10000. this is thfe
major bottleneck in my code, so i'd like to know if solving a tridiagonal
matrix is a parallelizable operation.
the only tridiagonal matrix algorithm i know of is the thomas algorithm.
the algorithm has a structure like:
get a result for a[i]
use a[i] to obtain a[i+1]
use a[i+1] to obtain a[i+2]
this looks decidedly non-parallelizable to me. i was wondering if anyone
knew of another algorithm which i could implement using MPI to get past this
(drop in code for fortran or C++ would be fantabulous!).
sorry for typos, but i just moved, and dsl hasn't kicked in yet. i'm on a
painfully slow ppp connection.
More information about the Beowulf