NPAC REU '92 Abstracts and Papers


Below are titles and abstracts of all technical papers from the 1992 Research Experiences for Undergraduates program in High Performance Computing, which was conducted during a 10-week period in Summer 1992 by the Northeast Parallel Architectures Center (NPAC) at Syracuse University. Links to hypertext and Postscript versions of the full papers are also provided.

The collected papers are available as the Journal of Undergraduate Research in High-Performance Computing, Volume 2, Bogucz, E.A., and Weinman, V.E., editors, Northeast Parallel Architectures Center at Syracuse University Technical Report SCCS-468, November 1992. Copies of this report are available via request to reu-info@npac.syr.edu.



The Generalized Hough Transform on an MIMD Machine

Daniel Baumann and Sanjay Ranka

Detecting objects in an image using the generalized Hough algorithm is a computationally intensive problem. An algorithm computing the generalized Hough transform for an MIMD architecture computer is developed. Additional techniques are implemented in an attempt to achieve greater speedups. Supportive experimental results on a CM-5 multicomputer are presented.

A Postscript version of the paper is available for immediate retrieval and viewing.


Parallel Implementation of a 2-D Convex Hull

Yvette Cordero and Uptal Roy

This paper addresses the determination of a two-dimensional convex hull on parallel computers. The calculation of a convex hull is important in a computational geometry approach to determining roundness error of a drilled hole. In addressing the problem of roundness error, this drilled hole can be represented by a simple polygon. Establishing the convex hull of this simple polygon is the starting point for assessing roundness error in this manufactured work piece. Several sequential algorithms already exist for finding a two-dimensional convex hull. Most of these algorithms use recursive programming techniques in order to obtain the convex hull. A parallel algorithm was developed from a divide-and-conquer approach to constructing convex hulls. The algorithm was written in CM Fortran and was implemented on the Connection Machine (CM-2). The parallel algorithm uses SCAN subroutines instead of recursive techniques to obtain the convex hull of a simple polygon.

A Postscript version of the paper is available for immediate retrieval and viewing.


An Efficient Parallel Implementation of an Automatic Test Pattern Generation Algorithm

Michael Harry and Dennis Shiau

This paper discusses an efficient parallel implementation of SIMPLE, an automatic test pattern generation algorithm. A new approach is presented in which processors that become idle during execution are reincorporated back into useful computation towards finding a test. The algorithm is designed for a MIMD-architecture computer and implemented on the CM-5. The approach is demonstrated to yield significant speedup over previous implementations.

A Postscript version of the paper is available for immediate retrieval and viewing.


Maximal Variety Graphs: A Model for Spacetime Structure

Ron Maimon and Lee Smolin

This paper suggests the need for a discrete space-time model and offers the graph as a possible solution. A proposed topology on the graph is examined and shown to be incomplete, leading to the standard metric on the graph to be accepted. The variety as a dynamical quantity is defined and numerically analyzed using high speed parallel machines. Graphs of maximal variety are shown to have or not to have desirable properties in terms of dimensionality, and are therefore appealing or to be discarded as possible spacetime structures.

Postscript version of the paper are available for immediate retrieval and viewing.


MIMD Implementation of Advanced Regional Prediction System

Nicole LaFontaine and Nancy McCracken

This project ported the Advanced Regional Prediction System version 3.0 from the University of Oklahoma CAPS group (Center for Analysis and Prediction of Storms) to an MIMD architecture, the Connection Machine 5. The CM-5 is a distributed-memory MIMD machine with a control processor and 32 processing nodes. The ARPS project does weather modeling by tracking the behavior of a three-dimensional disturbance in the atmosphere. The bulk of our research was finding an optimal mapping of the data structures from a sequential architecture with one processor to an MIMD architecture with 32 processors. Creating this mapping involved the design of a processor grid, routines to disperse the data structures onto the processor grid, and the development of message-passing features. Message-passing was important in this project because there were many ``nearest neighbor" calculations. In these types of calculations the value is computed based upon neighboring values. When a point needs a neighbor that is on a separate processor because of the mapping of the data structures, messages must be passed between the nodes. The program was done in FORTRAN with message-passing, using special library routines, and our preliminary results showed promise for a parallel message-passing code on the Connection Machine, Model CM-5

Postscript version of the paper is available for immediate retrieval and viewing.


Evaluation of Sorting Techniques on the CM-5

Paul D. Peters and Alok Choudhary

With the growing strain on database systems, it seems relevant to study the performance of parallel machines on database operations. One of the most important database operations is the sort operation. We tested the CM-5 with simulated databases (large arrays) using several sorting algorithms. The machine's performance was analyzed by observing sort times, sample times, and communication times. Experiments were done with different ``database'' sizes, and the number of processors used. The long-term goal is to evaluate the performance of the CM-5 on many database operations.

A Postscript version of the paper is available for immediate retrieval and viewing.


An Evaluation of Market Profitability for Four Stock Option Pricing Models

Neal Roodin and Kim Mills

A trading strategy is developed to evaluate the market profitability of four different pricing models in detecting under-priced call options. Included in the evaluation are the Black-Scholes Model and three binomial lattice models implemented on parallel computers. The results show that option pricing models that incorporate stochastic volatility have higher daily profitability and are therefore better able to detect under-priced call options than pricing models that assume constant volatility (e.g., the Black-Scholes Model). On an average, none of the four pricing models do significantly better at detecting under-priced call options than a random selection of call options. All five trading techniques are found to produce losses when used in simulated long-term trading.

Postscript version of the paper is available for immediate retrieval and viewing.


Evaluation of Simulated Tempering for Optimization Problems

Thomas Rose and Paul Coddington and Enzo Marinari

The application of massively parallel computers in simulating disordered physical systems and in the search for global minima of discrete optimization problems is addressed. Simulated tempering is a new algorithm that is presented as an alternative to the Monte Carlo procedure known as simulated annealing. A sequential version of this algorithm has been shown to improve greatly the simulation of a specific spin model as compared to conventional Monte Carlo techniques. In the present work, an efficient parallel program that performs simulated tempering on Ising spin models is produced and implemented on a MasPar system, using the MasPar Programming Language. The performance of the parallel version of the algorithm for finding the ground state of an Ising spin glass is then analyzed and compared to the performance of a parallel version of the simulated annealing algorithm. It is found that methods for tuning the variable parameters in the tempering algorithm must be discovered before simulated tempering can be as effective as simulated annealing for general optimization problems.

Postscript version of the paper is available for immediate retrieval and viewing.


Parallel Computation of Wavelet Transforms

Jennifer Rubbo and Jacques Lewalle

This work deals with the development of a parallel algorithm for performing signal processing using wavelet transforms. Wavelet transforms facilitate the recognition of patterns in irregular signals. The mathematical method involves the integration of the projection of the signal and a known function called a wavelet. The results are displayed graphically, each resulting value assigned a color. A parallel implementation of the wavelet transform was developed for the CM-2 in Fortran 90.

A Postscript version of the paper is available for immediate retrieval and viewing.


Fast Programs for Finding Maximum Planar Subgraphs

David G. Solt and Ernest Sibert

A previously reported, parallel method for calculating approximately maximum planar subgraphs has shown promising results, but was slow. The results of an initial implementation on the Connection Machine 2 prompted an investigation of more efficient methods of implementing the procedure. An improved implementation uses many fewer processors for the calculation, with substantially shorter communication times. By running several copies of the basic calculation simultaneously, the program performs one hundred times faster than earlier implementations.

Postscript version of the paper is available for immediate retrieval and viewing.


A Parallel Approach to Finding Prime Numbers and Visualizing their Distribution

{Pamela K. Speidel and Ernest Sibert

A data-parallel implementation of the Sieve of Eratosthenes has been used to calculate prime numbers up to approximately $2^{30}$ in C*. Alternative programming techniques to implement the sieve approach on the Connection Machine have been explored. In addition, ways to visualize the distribution of the prime numbers using the parallel graphics capability of the CM were investigated.

A Postscript version of the paper is available for immediate retrieval and viewing.


Q: A System for Doing Mathematics Really Fast by Computer

T.J. Willis and E. A. Bogucz

Environments for coupling symbolic computation with parallel numeric processing are demonstrated using mathematica and an nCUBE 2 multiple-instruction-thread-multiple-data-stream computer. The so-called ``Q'' environments provide programming models that allow distributed symbolic computing, as well as a bridge between sequential symbolic computing and distributed numerical computation. Scientific visualization, one strength of mathematica and a classical weakness of parallel machines, is showcased as a feature of the Q environments. Applications that demand parallel symbolic processing also are discussed.

A Postscript version of the paper is available for immediate retrieval and viewing.


Image Processing in Parallel: A Simple Model

Matthew Colby Young and Marek Podgorny and Tomacz Haupt

Image processing on a sequential machine is a tediously slow process. The need for image processing lies in scientific visualization, the entertainment industry, and a variety of real-world applications. Processing images is typically a very regular process. This paper outlines a model for using an SIMD-architecture computer to expedite the task. AVS, a front-end for a variety of visualization needs, is employed in this work. AVS is shown to provide a convenient system for parallel image processing, as well as providing an easy way to user interface.

A Postscript version of the paper is available for immediate retrieval and viewing.


REU Program, Northeast Parallel Architectures Center, Syracuse University, reu-staff@npac.syr.edu