PARALLEL IMPLEMENTATION OF 3+1-DIMENSIONAL RELATIVISTIC HYDRODYNAMICS

We discuss the parallel implementation of 3+1-dimensional relativistic hydrodynamics. The discretization of these equations is performed with a basis-spline collocation method. An iterative procedure for the time evolution will also be discussed. Approximately 60% of the computational effort is spent in performing vector-matrix multiplications that may easily be vectorized. Furthermore, for large lattices, these operations may be spread across the nodes of a parallel machine. We show how to accomplish this partitioning for the hydrodynamic evolution equations. A double-ring, message-passing algorithm is shown to enhance overall code performance when compared to a single ring on iPSC/i860 architectures. Overall code performance is discussed, and an application to relativistic nuclear collision physics is presented.