Parallel Scientific Computing

Work on this topic stems from a desire to exploit Linux clusters for AMR calculations. Optimization, performance modeling, RISC and modern memory architectures play a large role in code design, as do interconnet configurtions and specifications. C++ frameworks for AMR grid management are being evaluated for the next generation of reactive flow codes. Finally more efficient (i.e. less labor intensive) approaches to writing, extending and maintaining scientific codes too are investigated.

Load-balancers

Load-balancers are an active field of research in the quest for scalable flame simulations on block-structured adaptively refined meshes. The most current information on the load-balancers that are being developed can be found in Johan Steensland's load-balancer page

We have realized that there is no one load-balancer that works for all problems, all the time, especially since adaptive problems change their (parallel computing) characteristics as they evolve. The approach, therefore, is to characterize the problem at hand and choose and/or configure a load-balancer for a pre-existing collection of load-balancer. This "control system" approach to load-balancing, also called the meta-partitioner forms the bulk of our work. Some of our recent findings are documented in the papers below.

S. Chandra, M. Parashar and J. Ray, "Analyzing the Impact of Computational Heterogeneity on Runtime Performance of Parallel Scientific Components", Proceedings of the 15th High Performance Computing Symposium (HPC-07), SCS Spring Simulation Multiconference, Norfolk, VA, USA, March 2007.

Sumir Chandra, Manish Parashar and Jaideep Ray, "Dynamic Structured Partitioning of Parallel Scientific Applications with Pointwise Varying Workloads", Proceedings of the International Parallel and Distributed Processing Symposium, April 24-28, 2006. Rhodes, Greece.

J. Steensland and J. Ray, "A Partitioner-Centric Model for SAMR Partitioning Trade-Off Optimization : Part I," International Journal of High Performance Computing Applications, 2005, 19(4):409-422.

Johan Steensland and Jaideep Ray, "A Partitioner-Centric Model for SAMR Partitioning Trade-Off Optimization: Part II", In the proceedings of The 6th International Workshop on High Performance Scientific and Engineering Computing (HPSEC-04), held in conjunction with The 2004 International Conference On Parallel Processing (ICPP-04), in Montreal, Canada, Aug. 15-18, 2004.

Johan Steensland and Jaideep Ray, "A Partitioner-Centric Model for SAMR Partitioning Trade-Off Optimization: Part I", Proceedings of the 4th Annual Symposium of the Los Alamos Computer Science Institute (LACSI04). Paper distributed via CD-ROM, also here

Johan Steensland and Jaideep Ray, "A Heuristic Re-Mapping Algorithm Reducing Inter-Level Communication in SAMR Applications", Proceedings of the 15th IASTED International Conference on Parallel and Distributed Computing and Systems 2003 (PDCS03) . Paper distributed via CD-ROM, also here

Parallel software architecture

The other active field is in the use of the Common Component Architecture to design toolkits for the AMR simulation of flames . The CFRFS Toolkit is viewed as a collection of components, each embodying a functionality. Components can be chosen and "wired up" to create a simulation code. The aim here is demostrate that a component-based Toolkit can be computationally efficient (and parallel!) while being flexible enough to "absorb" the contributions of many collaborators, with diverse skills, which may not include advanced software development.

S. Lefantzi, J. Ray, B. A. Allan and H. N. Najm, "Computational Facility for Reacting Flow Science: Functionality, Reusability and Interoperability", Minisymposium on Numerical Software for Solving Problems in Computational Science and Engineering, SIAM Conference on Computational Science and Engineering, Feb 12-15, 2005, Orlando, FL.

Sophia Lefantzi and Jaideep Ray, "A Component-based Scientific Toolkit for Reacting Flows", Proceedings of the Second MIT Conference on Computational Fluid and Solid Mechanics, Boston, Mass. 2003.

S. Lefantzi, J. Ray and H. N. Najm, "Using the Common Component Architecture to Design High Performance Scientific Simulation Codes," Proceedings of the International Parallel and Distributed Processing Symposium, April 2003, Nice, France (distributed via a conference CD; also here)

Benjamin A. Allan, Robert C. Armstrong, Alicia P. Wolfe, Jaideep Ray, David E. Bernholdt and James A. Kohl, ``The CCA Core Specification in a Distributed Memory SPMD Framework,'' accepted, Concurrency : Practice and Experience , August 2001.

Created by Jaideep Ray