We present a novel scheme for the solution of linear differential equation systems on parallel computers. The Implicit Pipeline (ImP) method uses an implicit timeintegration scheme coupled with an iterative linear solver to solve the resulting differential algebraic system. The ImP method then allows for two independent mechanisms for parallelism: pipelining of the solution of several timesteps simultaneously, and pipelining of the successive linear iterations in the solution of each individual time-step. Since pipelining allows for a highly structured communication pattern, it is possible to achieve good parallel performance on large processor sets. Performance results from a Cray T3E are given.
John B. Pormann