next up previous contents
Next: Ising Model Evaluation Up: Applications and Performance Previous: Self-labeling   Contents


Graphical Simulation

An mpiJava application that performs a Potts Model Monte Carlo simulation using parallel Metropolis algorithms in two dimensions running on four processors is shown in Figure 6.6. One of the good feature of mpiJava is that users have full access to the Java AWT and other Java graphical packages to support GUI and visualize graphical output of application. Visualization of the collected data is a critical element in providing developers or educators with the needed insight into the system under study. In this Potts model simulation, the different colors represent the Q different spin states on a lattice. All the options of the simulation are customizable through a GUI. Processor $P_0$ handles all GUI input and visualization of output data. When users run simulation by clicking start button, input values are broadcast to other processors and then, after the processors generate a new configuration, processor $P_0$ gathers all updated data to plot the output in order to visualize the results. Measurements of standard quantities such as energy and magnetization are easily done in parallel by just calculating the quantities locally, sum over sites in each processor, and then combining the partial sums from all processors using MPI.Allreduce.

Figure 6.6: The Potts Model Monte Carlo Simulation.
\begin{figure}\centerline{\epsfig{figure=Figs/pottsdemo.ps,width=5.7in}}\end{figure}

In this demo, the total number of processors is broken up into a 2D grid, in X and Y directions. Users can also choose data distribution as row-, column- or block-wise by selecting the nodes in X direction option in GUI. This option is set after the number of running processors is given at the starting time by execution command as follows;

     prunjava 4 Potts
The prunjava script is a wrapper for the MPICH mpirun script. The first argument is the number of processors on which the program will be executed. The number of nodes in Y direction is set automatically by total number of running nodes and the number of nodes in X direction specified in the GUI. A list of available host computers may be given in an MPICH-style machines file in the local directory.

From this simulation, one is able to determine the average magnetization, the average energy, and the specific heat of the system.


next up previous contents
Next: Ising Model Evaluation Up: Applications and Performance Previous: Self-labeling   Contents
Bryan Carpenter 2004-06-09