APPLICATIONS: Computer & Information Sciences Advanced Networking Infrastructure & Research

Dynamic Load Balancing of Structured Adaptive Mesh Refinement (SAMR) Applications on Distributed Systems

A typical Adaptive Mesh Refinement (AMR) application may require a large amount of computing power. For example, a simulation of the galaxy formation requires a few days to execute on a 128-node SGI Origin2000 machine and requires more than 100GB of memory. Distributed systems provide an economical alternative to traditional parallel systems; however, the adaptive structure of AMR applications results in load imbalance among processors on distributed systems. Dynamic load balancing is an essential technique to solve this problem. ENZO is one of the successful parallel implementations of Structured Adaptive Mesh Refinement (SAMR) for use in astrophysics and cosmology. To sufficiently simulate the formation of galaxies, taking communication and latency issues into consideration, an estimated bandwidth of ~100Gbps would be required. Acknowledgment: The Globus Project; USA National Science Foundation; NASA; Alliance / National Center for Supercomputing Applications, University of Illinois at Urbana-Champaign, USA. Contact Zhiling Lan Department of Computer Science, Illinois Institute of Technology (IIT), USA lan@iit.edu Valerie Taylor Department of Electrical and Computer Engineering, Northwestern University (NU), USA taylor@ece.nwu.edu Collaborators Zhiling Lan, IIT, USA Valerie Taylor, NU, USA Greg Bryan, Nuclear and Astrophysics Laboratory, Oxford University, UK www.ece.nwu.edu/~zlan/research.html