15618 Project - Parallelized K-Means clustering

Build

Run make from the root folder. This will build a binary kmeans that is the main entrypoint for execution.

Benchmarks

The available options are:

• -in: path to input file
• -out: path to output file
• -k: number of cluster
• -n: number of data points
• -omp: run K-Means algorithm parallelized using OpenMP
• -mpi: run K-Means algorithm parallelized using MPI

Sample command:

$ ./kmeans -omp -k 32 -i 10 -in 16mil.csv -ce 16mil_centroids.csv -n 16000000
$ mpirun -n 32 kmeans -mpi -k 32 -i 10 -in 16mil.csv -ce 16mil_centroids.csv -n 16000000

Test data

Test data was generated with make_blobs. There generated test files used for performance benchmarking are:

• small.csv (1.6 million data points, 12 clusters)
• large.csv (16 million data points, 32 clusters)

Benchmarking and comparison

to be updated

Analysis

to be updated