Asynchronous Cellular Microphysiology Simulation

Pietro Cicotti, Scott Baden

Abstract

Simulations of dynamic environments are hard to parallelize efficiently, because the need to preserve causality exacerbates the impact of load imbalance. We discuss Tarragon, a runtime system that offers asynchronous data-driven execution, and supports relaxed synchronization and dynamic load balancing. Tarragon's API hides lower level system dependent details, taking advantage of any architectural features that can support its data driven execution model. We discuss recent results with a cell microphysiology simulator.