Parallel Heuristics for Bandwidth Reduction of Sparse Matrices with IBM SP2 and Cray T3D