tkphd / ppcproject

RPI CSCI-6360 Parallel Programming & Computing group project

PPCproject

RPI CSCI-6360 Parallel Programming & Computing 2014: Group Project

If you wish to contribute to this code, please fork this repository (http://github.com/fifthblackbird/PPCproject) into your GitHub account before cloning.

This is a group project intended to implement POSIX threads and to improve MPI-IO performance in MMSP (http://github.com/mesoscale/mmsp).

The code depends on the Mesoscale Microstructure Simulation Project (MMSP) libraries [1].

The source directory contains code for grain growth simulations using a sparse phase-field model [2] and a Potts Monte Carlo model [3].

Before building this code, download the MMSP source code and create symbolic links to MMSP/include and MMSP/algorithms.

E.g., if you store MMSP in ~/Downloads/mmsp, run

ln -s ~/Downloads/mmsp/include source/include
ln -s ~/Downloads/mmsp/algorithms source/algorithms

In source/, the Make commands default to the phase-field model: e.g., make or make parallel will build graingrowth.out or parallel_GG.out, respectively, both of which depend on graingrowth.cpp.

To compile the Monte Carlo version, append "mc" to the commands: e.g., make mc or make parallelmc will build graingrowth.out or parallel_MC.out, respectively, both of which depend on graingrowth_MC.cpp.

To build this code on an IBM Blue Gene/Q supercomputer, make sure the IBM xl_r compiler and zlib library are available. On the Rensselaer Polytechnic Institute BG/Q, AMOS, module load xl_r experimental/zlib. Then, make bgq or make bgqmc.

The recommended mode for this program is using --nonstop, which will intialize and coarsen a simulation domain. e.g., use

mpirun -np 6 ./parallel_GG.out --nonstop 3 voronoi.000.dat 200 200 4

to initialize and save a startup file (voronoi.000.dat) and a checkpoint file after 200 steps (voronoi.200.dat), using six MPI ranks with 4 pthreads each. Make sure you have hardware support for 24 threads before running this command!

Note that AMOS is big-endian. Most consumer PCs are little-endian. If you have downloaded a binary data file generated using MMSP on AMOS, make wrongendian and run it on the file. Afterwards, you can use standard MMSP utilities (e.g., mmsp2vti) to export a file for visualization.

References:

1. J. Gruber and T. Keller. The Mesoscale Microstructure Simulation Project. http://github.com/mesoscale/mmsp

2. I. Steinbach and F. Pezzolla. "A Generalized Field Method for Multiphase Transformations Using Interface Fields." Physica D 134 (1999) 385-393. DOI: 10.1016/S0167-2789(99)00129-3

3. S.A. Wright, S.J. Plimpton, T.P. Swiler, R.M. Fye, M.F. Young, and E.A. Holm. "Potts-model grain growth simulations: Parallel algorithms and applications." SAND Report (1997) 1925.