Arrays that are distributed in a block-cyclic fashion are important for many applications in the computational sciences since they often lead to parallel algorithms with good load balancing properties. We consider the problem of redistributing such an array to a new block size. This operation is directly expressible in High Performance Fortran (HPF) and will arise in applications written in this language. Efficient message passing algorithms are given for the redistribution operation, expressed in the standardized message passing interface, MPI. The algorithms are analyzed and performance results from the IBM SP-1 and Intel Paragon are given and discussed. The results show that redistribution can be done in time comparable to other collective communication operations, such as broadcast and MPI_ALLTOALL.

Abstract Many large‐scale engineering and scientific calculations involve repeated updating of variables on an unstructured mesh. To do these types of computations on distributed memory parallel computers, it is necessary to partition the mesh among the processors so that the load balance is maximized and interprocessor communication time is minimized. This can be approximated by the problem of partitioning a graph so as to obtain a minimum cut, a well‐studied combinatorial optimization problem. Graph partitioning is NP complete, so for real world applications one resorts to heuristics, i.e. algorithms that give good but not necessarily optimum solutions. These algorithms include recursive spectral bisection, local search methods such as Kernighan‐Lin, and more general purpose methods such as simulated annealing. We show that a general procedure enables us to combine simulating annealing with Kernighan‐Lin. The resulting algorithm is both very fast and extremely effective.

Adaptive load distribution is necessary for parallel applications to co-exist effectively with other jobs in a network of shared, heterogeneous workstations. We present three methods that provide such support for PVM applications. Two of these methods, MPVM (migratable PVM) and UPVM (user-level PVM), adapt to changes in the workstation environment by transparently migrating the virtual processors (VPs) of the parallel application. A VP in MPVM is a Unix process, while UPVM defines lightweight process-like VPs. The third method, ADM (adaptive data movement), is a programming methodology for writing programs that perform adaptive load distribution through data movement. These methods are discussed and compared in terms of effectiveness, usability and performance.>