Abstract approved:

optimizations for distributed shared-memory machines, although the same techniques can be used for sequential machines with a memory hierarchy. Our compiler optimizations are based on an algebraic representation of data mappings and a new data locality model. We present a pure data transformation algorithm

class can be solved independently on each processor requiring no synchronization. However, dynamic inter-class and intraclass load balancing must be exploited to ensure that each processor gets an equal amount of work. Experiments on a 12 processor SGI Origin 2000 shared memory system show good speedup

on the variety of parallel computers that can support the paradigm, including shared memory, distributed memory, and shared memory array multiprocessors. This paper compares the SGI Power Challenge, a shared memory multiprocessor, with the Intel Paragon, a distributed memory machine. This paper addresses two

The construction of a cache-coherent distributed shared memory (DSM) machine involves many organizational and implementation trade-offs. This paper studies the performance implications of these trade-offs as made on some real DSM machines. We focus on characteristics related to communication

In this paper we describe a new parallel iterative technique to solve a set of linear equations. The technique can be applied to any serial iterative scheme and involves pipelining sllccessive iterations. We give an example of this technique by modifying the classical successive Qver-relaxation method (SOR). The algorithm is implemented on a Sequent Balance 21000 and the experimental results are presented.

In this paper, we present CGMgraph, the first integrated library of parallel graph methods for PC clusters based on Coarse Grained Multicomputer (CGM) algorithms. CGMgraph implements parallel methods for various graph problems. Our implementations of deterministic list ranking, Euler tour, connected components, spanning forest, and bipartite graph detection are, to our knowledge, the first efficient implementations for PC clusters. Our library also includes CGMlib, a library of basic CGM tools such as sorting, prefix sum, one-to-all broadcast, all-to-one gather, h-Relation, all-to-all broadcast, array balancing, and CGM partitioning. Both libraries are available for download at

class can be solved independently on each processor requiring no synchronization. However, dynamic inter-class and intraclass load balancing must be exploited to ensure that each processor gets an equal amount of work. Experiments on a 12 processor SGI Origin 2000 shared memory system show good speedup

Abstract. There are several alternative MPI implementations available to parallel application developers. LAM MPI and MPICH are the most common. System vendors also provide their own implementations of MPI. Each version of MPI has options that can be tuned to best t the characteristics of the application and platform. The parallel application developer needs to know which implementation and options are best suited to the problem and platform at hand. In this study the RTCOMM1 communication benchmark from the Real Time Parallel Benchmark Suite is used to collect performance data on several MPI implementations for a Sun Enterprise 4500. This benchmark provides the data needed to create a re ned cost model for each MPI implementation and to produce visualizations of those models. In addition, this benchmark provides best, worst, and typical message passing performance data which is of particular interest to real-time parallel programmers. 1

Parallel programs written in MPI have been widely used for developing high-performance applications on various platforms. Because of a restriction of the MPI computation model, conventional MPI implementations on shared memory machines map each MPI node to an OS process, which can suffer serious