Supercomputing pp 926-938 | Cite as

# Parallel matrix factorizations on a shared memory MIMD computer

## Abstract

This paper is concerned with the study of parallel algorithms for matrix factorization on a shared memory multiprocessor MIMD type computer. We consider the implementation of LU and WZ factorizations of general nonsymmetric matrices when the number of processors p is ∼O(n), where n is the order of the matrix. We show how each of these methods can be divided into noninterfering tasks which can then be executed in parallel. By studying the precedence graph of these tasks we are able to find a schedule for each algorithm which is optimum for a certain number of processors. We also consider the use of the resulting factors to solve a linear system of equations and compare the two algorithms in terms of their speedup and efficiency. It is shown that the parallel WZ algorithm attains a better efficiency using only half the processors of Doolittle's method.

## Keywords

Completion Time Parallel Algorithm Shared Memory Precedence Constraint Task System## Preview

Unable to display preview. Download preview PDF.

## References

- 1.E. G. Jr. Coffman and P. J. Denning, Operating systems theory, Prentice-Hall, Englewood Cliffs, N.J. (1973).Google Scholar
- 2.D. Evans and M. Hatzopoulos, A parallel linear system solver,
*Inter. J. Computer Math.*, 7, 227–238, 1979.Google Scholar - 3.D. Evans and A. Hadjidimos, A modification of the quadrant interlocking factorization parallel method,
*Inter. J. Computer Math.*, 8, 149–166, 1980.Google Scholar - 4.G. A. Geist and M. T. Heath, Matrix Factorization on a hypercube multiprocessor,
*Hypercube Multiprocessors 1986*, Proceedings of the first Conference on Hypercube Multiprocessors, 161–180, 1986.Google Scholar - 5.A. George, M. T. Heath and J. Liu, Parallel Cholesky factorization on a shared-memory multiprocessor,
*Linear Algebra and Its Appl.*, 77, 165–187, 1986.Google Scholar - 6.M. Hatzopoulos, A symmetric parallel linear system solver,
*Inter. J. Computer Math.*, 13, 133–141, 1983.Google Scholar - 7.M. Hatzopoulos and N. M. Missirlis, Advantages for solving linear systems in an asynchronous environment,
*J. of Comp. and Appl. Math.*, 12 & 13, 331–340, 1985.Google Scholar - 8.D. Heller, A survey of parallel algorithms in numerical linear algebra,
*SIAM Rev.*, 20, 740–777, 1978.Google Scholar - 9.S. P. Kumar, Parallel algorithms for solving linear systems of equations on an MIMD computer, Ph.D, Department of Computer Science, Washington State University, 1982.Google Scholar
- 10.H. T. Kung, Synchronized and asynchronous paralllel algorithms for multiprocessors,
*Algorithms and Complexity: New Directions and recent results*, Academic Press, New York, 153–200, 1976.Google Scholar - 11.R. E. Lord, J. S. Kowalik and S. P. Kumar, Solving linear algebraic equations on an MIMD computer,
*J. Assoc. Comput. Mach.*, 30, 103–117, 1983.Google Scholar - 12.N. M. Missirlis, Scheduling parallel iterative methods on multiprocessor systems,
*Parallel Computing*(to appear).Google Scholar - 13.A. Sameh and D. Kuck, On stable parallel linear systems solvers,
*J. ACM*, 25, 81–91, 1978.Google Scholar - 14.C. L. Seitz, The cosmic cube,
*Comm. ACM*, 28, 22–33, 1986.Google Scholar - 15.J. Shanehchi and D. Evans, New variants of the quadrant interlocking factorization (QIF) method,
*CONPAR 81 Conf. Proc. Lecture Notes in Computer Science IIII*, W. Handler, Springer-Verlag, Berlin, 493–507, 1981.Google Scholar - 16.R Voigt, The influence of vector computer architecture on numerical algorithms,
*High speed computer and algorithm organization*, Academic Press, New York, 229–244, 1977.Google Scholar