Applicability of Load Balancing Strategies to Data-Parallel Embedded Runge-Kutta Integrators

摘要

Embedded Runge-Kutta methods are among the most popular methods for the solution of non-stiff initial value problems of ordinary differential equations (ODEs). We investigate the use of load balancing strategies in a data-parallel implementation of embedded Runge-Kutta integrators. Since the parallelism contained in the function evaluation of the ODE system is typically very fine-grained, our aim is to find out whether the employment of load balancing strategies can be profitable in spite of the additional overhead they involve.