Out-of-core mergesort

This is the implementation of the mergesort algorithm in an out-of-core manner using FastFlow,OpenMP and MPI.

Project files organization

The source for all the implementations is contained in the src directory:

• core contains the core classes used in all the implementations,specifically
  • sequential contains ms_sequential.cpp which is the sequential version.
  • ff_singlenode contains ms_ff_singlenode.cpp which is the single-node version done with FastFlow. The file contains allso the FastFlow's building blocks used in the implementation.
  • openmp_singlenode contains ms_openmp_singlenode.cpp which is the single-node version done with OpenMP.
  • multinode contains ms_multinode.cpp which is the single-node version done with OpenMP+MPI.

Additionally we have:

• The file utilities/payload_generator.cpp which which hold the utility to generate the files used in the tests and check their sorting/validate the data.
• The folder benchmarks which hold the bash files used to benchmark/debug the various implementations
  • the files with suffix _cluster.sh are used to do the tests on the cluster the other ones are used for local debugging.

Compiling the project

In the Makefile we have the targets for each version:

• ms_sequential for the sequential version.
• ff_singlnode for FastFlow's single-node version.
• openmp_singlenode for OpenMP's single-node version.
• mpi_multinode for the OpenMP+MPI multi-node version.

The additional payload_generator target will generate the corresponding payload generator utility.

Running the project

All versions use the same arguments, this is the full command:

• <your version> -t <number of threads> -v (1 if you want to see want to plot values after sorting 0 if not) -i <input file name> -o <output file name> -m <maximum buffer memory limit in bytes>

All parameters are mandatory except for the -t parameter in the sequential version which is not being used.

If you want to use the payload generator the arguments are:

• payload_generator -p <max payload size in bytes> -r <number of records> -o <output file name> -d For this the -d flag is used to check if the output file is sorted and has len values above 8.

running the benchmarks

For the benchmarks you can just run the bash scripts in the benchmark directory corresponding to the version you want to try, each benchmark will take as a first argument a file where it will store the results and will display all outputs from the current experiment in the console.