Performance analysis of direct N-body algorithms on highly distributed systems

We present a performance analysis of different parallelization schemes for direct codes used in the simulation of astrophysical stellar systems. These codes compute the gravitational interaction among stars in an exact way and have a computational complexity of O(N^2). Significant improvement in the performance of direct N-body codes can be obtained by means of general purpose massively parallel supercomputers and of special purpose computers like GRAPE hardware. We compare the performance of parallel algorithms on different architectures including a cluster, a supercomputer and two computational grids. The best performance is obtained in combination with GRAPE-6 hardware but highly distributed computational Grids also appear very promising. The simulation of a globular cluster containing about one million stars, currently one of the most challenging numerical problems in astrophysics, is feasible only in combination with GRAPE hardware or on massively parallel architectures