alexgrigoras/distributed_sum

Distributed application of computing the sum of a matrix

Distributed sum

Description

Distributed application of computing the sum of a matrix:

• Start with one root process (p0) and a matrix of integers of dimension m lines by n columns;
• The matrix is dynamically generated using random numbers;
• The m and n parameters are provided as program arguments;
• The goal is to obtain for each line i (i=1..m) the sum (si) of its elements;
• The root process starts by spawning m child processes (p1...pm). For each child process, it sends the corresponding line in the matrix (li) as an array of n elements. A child process, computes the sum for the input array and send it back to the parent process;
• Depending on the size of the received array, a child process does the following:
  • if the array size is less than k (ex: 1000) elements, the child process computes the sum by itself and sends the result back to its parent;
  • otherwise, the child process splits the array in half, spawns two children of its own (thus becoming itself a parent), sends each half to the corresponding child and waits for the results. Upon receiving the partial sum results, it sums them and sends the result back to its parent.

Implementation

Implemented in C using MPI.

Run using the command (on Linux):

mpirun -np 1 distributed_sum_master [m] [n] [k]

License

MIT License.