Alexander Kissinger PHYS244 Spring 2013
Intro: There are three implementations that can be ran in from source: C-MPI, Serial- GPU, and Hybrid-MPI versions. *NOTE: This was tested on and developed to be supported on, primarily, a Raspberry Pi Cluster. Compiling on any other system will not guarentee properly functioning code. This README also assumes that you have a Raspberry Pi cluster in your disposal.
Preparing the code: Before running/compiling, here are the following dependencies: python-dev python-imaging (PIL) mpi-defualt-bin (open-mpi) mpi-defualt-dev mpi4py numpy
The above were installed on a Debian Flavored Linux Distribution (Raspbian) using Aptitude.
After installing the dependencies, you will have to go into the C_MPI_Implementation folder and perform a 'make' before making making any calls to the C-MPI simulation.
The run.py file is the head file for running the simulations and should be used as so. Of course you can always study the directories and make MPIrun calls on your own. The following illestrates how to perform the simulations:
To run a C_MPI simulation run the following: ./run.py cmpi hostfile length_of_side num_evolutions
To run a serial-GPU simulation run the following: ./run.py serial hostfile length_of_side num_evolutions
To run a hybrid-MPI simulation run the following: ./run.py hybridmpi hostfile length_of_side num_evolutions
All simulations will output the time it takes for each task to compute the matrix of length provided at the number of evolutions provided. Be advised that the current implementation limits inputing different sized lengths for the width and height and simply accepts a single argument and squares it for the area. This is becuase the base Conway.py and OpenGL ES is very selective on the problem set size, as well as the input texture, and until we have streamlined this issue we will wait before providing this feature until a later release.
In the case that either the Hybrid or Serial simualtions crash, it will most likely be a result of picking length parameters that are to high/low or ones that the OpenGL shader does not like. If this happens just choose different parameters.
C-MPI implementation was found on Shodor's main site and belongs to Samuel Leeman-Munk. The Hybrid-MPI was developed by Alexander Kissinger, guided by Rick Wagner (SDSC), and belongs to the San Diego Super Computing Sandbox Project. The Hybrid-MPI code, as well as the Serial-GPU implementation, was based-off, and originally developed by, Pi3d (pi3d.github.io) and the Pi3d Team (a special thanks to Paddy).