Math 627 - Introduction to Parallel Computing
Spring 2010 - Matthias K. Gobbert
Presentations of the Class Projects

  1. Wednesday, May 12, 2010, 01:05-01:20
    Creating an Interface between MPI and Java
    Ben Harris, Department of Computer Science and Electrical Engineering
    There are currently no other easily obtainable interfaces for MPI and Java. The following describes an interface that was developed to allow a program to access the MPI library from within a Java program. In general, a program using this library written in Java was multiple times slower than an equivalent program written in C.

  2. Wednesday, May 12, 2010, 01:05-01:20
    Improved Input Parameter Distribution to Parallel Processes in Support of Operator Upscaling for the Acoustic Wave Equation
    John Burghardt, Department of Mathematics and Statistics
    The purpose of this paper is to report on the best method to read in large data files containing input parameters for the operator upscaling code for the acoustic wave equation. The upscaling code is written in C with MPI for parallel processing. Six new methods will speed up and simplify the code while using less memory. However, there is little difference in runtime between the new methods. This leads to the conclusion that when setting up input parameters, code should be simple and flexible with memory usage. Testing will be performed on the tara cluster, which is maintained by the UMBC High Performance Computing Facility (www.umbc.edu/hpcf). The tara cluster has a total of 86 nodes, each of which has two quad core Intel Nehalem X5550 processors (2.66 GHz, 8192 kB cache), 24 GB of memory and a 120 GB local hard drive. A high performance InfiniBand interconnect connects the 82 compute nodes, 2 development nodes, management node, front end node and the 160 TB of central storage.

  3. Friday, May 21, 2010, 01:05-01:20
    A Parallel Performance Study for Long-Time Simulations of Calcium Waves in a Heart Cell
    David W. Trott, Department of Mathematics and Statistics
    Calcium flow within a single heart cell can be modeled by a system of coupled, time-dependent reaction-diffusion equations with no-flux boundary conditions and a given set of initial conditions. In this talk, we present performance and scalability studies on a parallelized numerical code for the full three-species application problem utilizing the High Performance Computing Facility's (HPCF) new 86-node distributed-memory cluster tara. It will be shown that increasing the number of processes used for large problems significantly reduces the wall clock time needed to carry out the simulation. Using the performance study, we can formulate a rationale for the most effective usage of node resources for production runs of the full application code.

  4. Friday, May 21, 2010, 01:25-01:40
    A Perl Platform Embedded in C for Parallel Execution of a Monte Carlo Study on Protein Similarity
    I. Lee Nolen, Department of Mathematics and Statistics
    A Perl interpreter is embedded in C as a parallel platform for execution of Monte Carlo studies on the similarity of pairs of protein families. The interpreter provides an interface for serial Perl code to be executed during run-time in conjunction with MPI functions in C. The platform is built for trending of results for any Monte Carlo study of a particular design with minimal code changes. Limitations imposed by Perl is identified and a design paradigm is described for use of the interface of C based MPI and Perl. Assessment of speed-up is inconclusive while notably increased throughput of data is verified.

  5. Friday, May 21, 2010, 01:45-02:00
    ADI As Preconditioner in 3-D
    Kyle Stern, Department of Mathematics and Statistics
    The alternating directions implicit (ADI) method is a classical iterative method for numerically solving linear systems arising from discretizations of partial differential equations. We use ADI as a preconditioner for Krylov subspace methods for a linear system from a finite difference approximation of an elliptic test problem in three dimensions. The method is attractive because it allows highly efficient matrix-free implementations both in serial MATLAB and parallel C with MPI.

Copyright © 2001-2010 by Matthias K. Gobbert. All Rights Reserved.
This page version 1.0, May 2010.