Efficient solution of the eigenproblem of banded symmetric Toeplitz matrices on multicomputers

In this paper we study the parallelization of the bisection method to compute the eigenstructure of banded symmetric Toeplitz matrices. We design and implement four efficient parallel algorithms on a distributed memory multiprocessor based on a network of transputers. One of the parallel algorithms uses a static distribution of the calculations and the three remaining algorithms use a farming technique to dynamically balance the load. We also present in this paper a complete analysis of the experimental performances of the algorithms and we compare their computational results when we modify the bandwidth and the matrix size.<>