Skip to main content
U.S. Department of Energy
Office of Scientific and Technical Information

A portable parallel programs library for simulation and optimization in multiprocessor systems

Thesis/Dissertation ·
OSTI ID:6037536
The availability of more and more cost-effective and powerful parallel computers has enhanced the ability of the operations research community to solve more laborious computational problems. Most of the research on numerical algorithms using parallel computing is related to systems of linear equations, partial differential equations, the fast Fourier transform, and linear algebra such as matrix multiplication. Only rarely have parallelization efforts been aimed at those techniques in which operations researchers are interested: algorithms for mathematical programming, integer programming, dynamic programming, and simulation runs dispatcher. This dissertation develops a framework for the design and implementation of a transportable operations research library in C for use on several shared memory multiprocessor systems, such as Encore Multimax, Alliant FX/8 and Sequent Balance machines. A non-gradient unconstrained nonlinear optimization algorithm and a simulation runs dispatcher are parallelized to serve as examples of the feasibility of this framework. The analysis of the covariance has detected the importance of the system load factor, which has been ignored in most of the other research. Nearly linear speedup for the parallel runs dispatcher suggests that the library, if used properly, can reduce the execution time of a program in almost n times if n processors are available. The existence of a portable parallel operations research library can relieve most of the operations researchers from doing parallel programming, and can serve as a benchmark for the performance evaluation of multiprocessor systems.
Research Organization:
Iowa Univ., Iowa City, IA (USA)
OSTI ID:
6037536
Country of Publication:
United States
Language:
English