Parallel implementation of Gauss-Seidel algorithm used to solve linear system of equations using MPI (Message Passing Interface). Aditional information about the method on “Parallel Computer Architecture. A Hardware/Software Approach. David Culler, Jaswinder Pal Singh, Anoop Gupta”.
The program automatically creates a nxn matrix and initialices it with random numbers then solves this matrix using a parallel implementaion of Gauss-Seidel algorithm.
Note: some comments and parts of the code are written in Spanish.
Inputs
n: positive integer number representing the matrix size to be create (different of the-ninput ofmpiexeccommand)communication: either 0 or 1 for point-to-point communication or collective communication
Example
To create a 1026x1026 matrix and solve it using collective communication with 4 processes the command will look like this:
mpiexec –n 4 –f machine_file ./gs_mpi 1026 1
Output
At the end of the execution some interesting information is printed. The information covers the following data:
Matrix size: matrix sizeCommunication: communication type# procesos: total number of processesTIEMPO SOLUCION: time to convergetiempo mem. and MPI: time loading matrix and MPI stuffstiempo scatter matriz: time consumed for scattering matrixtiempo gather matriz: time consumed for gathering matrixtiempo en solver: time spent in solvingtiempo total comu. diff y done: time consumed to communicate some variablestiempo total comu. filas vecinas: time consumed communicating rows between processesiteraciones: total number of iterationst. media por itera.: mean time per iterationt. media comu. diff y done: mean time communicating some variblest. media comu. filas vecinas: mean time spent for communicating rows between processes
This code was part of my studies and I hope it can be useful for you in your studies on Parallel Computing. This piece of code can be used for testing performance on different computers and other some interesting stuff you can imagine.