ryant26 / ece420lab4 Goto Github PK

• Need all stats for a single computer and cluster
• Follow same testing procedure and display as all previous labs
  • 5 runs at each number of threads (processes in this case) take average
  • generate speedup and efficiency charts
• Finally put all the raw data, charts and analysis in the report

This goes with #2

The way we have worked with matrices in this class so far wont work well with MPI's communication functionality. We need to store the matrix in row major form (a way to store a matrix in 1 contiguous block of memory, that is, a single array).

The supporting operations we need for this are:

double get_element(int i, int j);
void set_element(int i, int j, double value);

// The matrix argument would be just a pointer
// That is, it has not already been malloc'd
// mallocing the memory would be the job of create_matrix
void create_matrix(double ***matrix, int height, int width);

The utility to read in matrices would have to be adapted to use this.

An easy approach would be :

• Broadcast the entire weight matrix to every process
  • The last column in the weight matrix needs to contain the number of outgoing edges for that node
  • IE Weightmatrix[n][i] contains the number of outgoing edges from node i
• Each process calculates the page rank for a subset of nodes (the size of the subset depends on the number of processes)
• The master process gathers all the page ranks for that iteration
• repeat.

• Clean out unnecessary files
• Make sure makefile conforms to lab spec

@jbcuglietta: please let me know when your testing is finished and I will create this.