Data Distribution Models and Algorithms

Data distribution, and its interaction with parallelism and load balancing, is the key unsolved problem for compiling for parallelism for distributed memory computers. Many diierent techniques and algorithms have been proposed or implemented, suited to diierent programming environments, target architectures, and applications. However , there is little uniformity, common models, or obvious approach to classifying and comparing such techniques and algorithms. This paper provides a general framework for comparing diierent methods based upon their generality and underlying models. We show the equivalence of several methods, discuss the major unsolved problems and suggest avenues that might lead to solutions.