Parallel Tree N-body Code: Data Distribution and DLB on the CRAY T3D for Large Simulations

We describe a strategy for optimal memory and work distribution. We have performed a series of tests to find an optimal data distribution in the Cray T3D memory, and to identify a strategy for the Dynamic Load Balance (DLB). The results of tests show that the step duration depends on two main factors: the data locality and the network contention. In a very large simulation, due to network contention, an unbalanced load arises. To remedy this we have devised an automatic work redistribution mechanism which provided a good DLB.