DAG SCHEDULING ALGORITHMS FOR ENTITY-LEVEL SIMULATIONS

The continuing deployment of high-performance network technology enables the development of computing platforms that aggregate widely distribute hardware resources. The vision for such a Computational Gridpromises computational platforms of unprecedented power for scientific applications. However, application developers need to rethink implementation paradigms in order to realize this potential. In this paper, we identify a class of increasingly important applications, entity-level simulations , which currently cannot use largescale computing platforms effectively. We will show how careful application-aware scheduling can enable such applications to utilize large distributed heterogeneous platforms. Our initial approach is to exploit the structure of entity-level applications and leverage existing Directed Acyclic Graph (DAG) scheduling techniques. We validate our approach by simulating a realistic application scenario on several synthetic platforms, including a representative Computational Grid testbed.