Developed by Dimitrios Karageorgiou,
during the course Parallel And Distributed Systems,
Aristotle University Of Thessaloniki, Greece,
2017-2018.

The provided implementation is a hybrid classifier, that utilizes both MPI communications for communications between cluster's nodes and OpenMP for parallel processing in the shared memory environment of each node.

Its original target was HellasGrid nodes, though it may run upon any cluster and/or setup that provides at least an MPI 1.0 implementation.

make all

In order to compile an implementation of MPI 1.0 or above, like OpenMPI, should be present. Also, a compiler that supports OpenMP and C99 is required.

export OMP_NUM_THREADS=<threads_per_process>
make run p=<num_of_processes>

or

export OMP_NUM_THREADS=<threads_per_process>
make runp p=<num_of_processes> data=<path_to_datafile> labels=<path_to_labelfile> k=<nearest_neighhbors_num> [results=<path_to_precalculated_results_file> [indexes=<path_to_precalculacted_indexes_file>]]

where arguments in [] are optional.

export OMP_NUM_THREADS=<threads_per_process>
make run_blocking p=<num_of_processes>

or

export OMP_NUM_THREADS=<threads_per_process>
make runp_blocking p=<num_of_processes> data=<path_to_datafile> labels=<path_to_labelfile> k=<nearest_neighhbors_num> [results=<path_to_precalculated_results_file> [indexes=<path_to_precalculacted_indexes_file>]]

where arguments in [] are optional.

For clusters that support qsub and I2G_MPI_START mechanism, there is a testing script under helper_scripts/local_cluster_submit named local_submit.sh.

For submission upon HellasGrid (and further EGI) infrastracture, there is a JDL file and two required scripts under helper_scripts/hellasgrid_submit.