Math 627 - Introduction to Parallel Computing
Spring 2012 - Matthias K. Gobbert
Presentations of the Class Projects

Friday, December 14, 2012, 01:00 p.m.

  1. 01:00-01:15
    Performance Study on Database Search for Linkage Path Generation Problem on Tara Cluster
    Cong Yue, Department of Mechanical Engineering
    This presentation presents a performance study for searching the database, which is used to solve linkage design for path generation problem. The goal is to design a fast linkage design system that can solve linkage path generation problem. Path generation problem aims at find a linkage of which a point called "coupler point" can trace approximately to the task curve. To solve this problem, a database of Fourier descriptors is built firstly, that characterize the curves of planar linkages. Then, search algorithm is developed to compare the Fourier descriptors of the task path against those of the linkage samples in the database and retrieves a design candidate. In order to achieve fast on-line search, this paper presents a parallel algorithm on database search and compares it with serial approach Database with samples vary from 100,000 to 10,000,000 is tested on the tara cluster, which is maintained by the UMBC High Performance Computing Facility (www.umbc.edu/hpcf), and the result shows that this parallel search approach can be applied to our linkage design system, and the search speed is improved dramatically compared to former serial search approach.

  2. 01:15-01:30
    Performance Study of Ray Tracer Algorithm Using MPI on the Cluster tara
    Islem Long, Department of Mechanical Engineering
    We numerically model human red blood cell deformation via Optical Stretchers to understand the effect the malaria parasite has on the mechanical properties of the cell. This paper presents an attempt at parallelizing the ray-tracer algorithm used by the Dynamic Ray Tracing method. Because we use a simpler version of the ray tracer algorithm, the code as it is did not benefit from being parallelized. This work is accomplished thanks to Dr. Charles D. Eggleton from the Mechanical Engineering Department, Dr. Matthias Gobbert from the Mathematical and Statistics Department, HPCF and the grant from NIH.

  3. 01:30-01:45
    Evaluating the Diehard Test Battery
    Jonathan McHenry, Department of Mathematics and Statistics
    The importance of selecting a good random number generator (RNG) continues to increase as high performance computing raises the scale of stochastic simulations and MCMC statistical methods. A good RNG should not have non-random patterns that are easily detected by standard tests of randomness. We evaluate the Diehard test on a few standard RNGs as well as the first few (25 million) digits of pi, which are conjectured to be equidistributed in any base. Ultimately, no comparison of these RNGs can be drawn from the Diehard tests - they all pass all of the Diehard tests. Diehard has failed in the sense that it cannot discriminate between these generators. This work is collaborative with Dr. Matthias Gobbert and Dr. Nagaraj Neerchal from the Department of Mathematics and Statistics.

  4. 01:45-02:00
    Parameter Sensitivity Analysis of Stochastic Chemical Reaction System Via Parallel RPD method
    Ting Wang, Department of Mathematics and Statistics
    Parameter sensitivity analysis is a powerful tool in the building and analysis of chemical network models. Sensitivity analysis of reaction system using probabilistic approach to characterize the randomness of the system is commonly performed using Monte Carlo methods which requires large numbers of simulation to generate accurate statistics. The regularized pathwise derivative (RPD) method that was developed by Muruhan Rathinam et al. (2012) and uses the random time change (RTC) representation of stochastic chemical kinetics gives an efficient way for sensitivity analysis. In this report, the Mersenne Twister random number generator will be used to generator random sequence. Also,the parallel version RPD algorithm will be implemented for reversible isomerization model. Moreover, the speedup and efficiency of the parallel RPD algorithm will be provided.

  5. 02:00-02:15
    Performance Study of Calcium Waves Simulation with Finite Volume Method
    Xuan Huang, Department of Mathematics and Statistics
    Spontaneous calcium sparks can lead to propagation of a self-initiated calcium wave under certain conditions in a heart cell. A model for diffusion waves of calcium ions in a heart cell is given by a system of coupled, time-dependent reaction-diffusion equations. The scheme is based on a Finite Volume discretization in space and uses the implicit numerical differentiation formulas as time integrator. A performance study by long time simulations of calcium flow is given, with suggestions made for number of processes to be used in future tests. The results presented here were obtained via simulations on the cluster tara in the UMBC High Performance Computing Facility and in collaboration with Dr. Matthias K. Gobbert at UMBC and researchers at the University of Kassel, Germany.

Copyright © 2001-2012 by Matthias K. Gobbert. All Rights Reserved.
This page version 1.0, December 2012.