|Program for Research in Computing and Information Sciences and Engineering|
PARALLEL AND DISTRIBUTED COMPUTING GROUP
The Parallel and Distributed Computing Group performs research in the design, implementation, and efficiency measurements of parallel algorithms. It also addresses research issues related to parallel and distributed computing systems with an emphasis in high-performance cluster computing and Grid computing. Our work includes a wide spectrum of experiences from computing systems (e.g., self-optimizing software, hybrid multithreaded-distributed computing environments, performance testing, etc), to mathematical modeling and simulation of physical and biological phenomena, and the mathematics of algorithm design and optimization. http://www.ece.uprm.edu/PDC
Current Participation in Competitive Research Grants
Strategic R&D Alliances with other Institutions
Supercomputing Center (Dr. David Deerfield, Biomedical Division).
Domain Decomposition Methods – Prof. Wilson Rivera
A new method based on explicit predictor and implicit corrector (EPIC) for generating numerical interface boundary conditions between subdomains has been proposed. We have applied the EPIC method to the solution of nonlinear equations on structured grids. The numerical results have shown that for transient problems the boundary treatment performed using the EPIC method yields significant improvement in accuracy compared to the traditional time-lagging (TL) method. The EPIC method showed better accuracy than time-lagging for detecting unsteady shocks. Also, the EPIC method demonstrated high quality solution at high CFL conditions, while the TL method demonstrated a reduction in quality as the CFL number was increased. In addition, the results showed that the new parallel algorithm is scalable as the number of processors increases. We plan to extend the EPIC methodology to unstructured grids. Unstructured grids offer a means to easily generate grids around very complex configurations since no predefined connectivity must be maintained, therefore, providing more flexibility and more optimal meshing point distribution. Strategies of treating subdomain interface connectivity and coupling in the parallel unstructured solution algorithm are being investigated.
Special Purpose Fast Fourier Transform (FFT) Algorithm - Prof. Jaime Seguel
The FFT is crucial to scientific computation. In several of these instances, the data to be transformed possesses special features such as symmetries, irregular shapes, or, as in physics, some degree of accuracy is demanded. The proposed research is intended to design and test algorithms for applications such as Volterra filters, Poisson solvers, and Crystallographic FFTs. This work concentrates on the use of mathematical properties for improving the computation of multidimensional FFTs of data sets endowed with special features such as symmetries or irregular shape, and improving the precision of FFT computations. The work contemplates implementations on parallel computing environments as well as the design of special purpose compilers for the efficient production of performance critical code segments.
Self-optimization Methodologies for Local Area Networks and the Grid – Prof. Jaime Seguel & Wilson Rivera.
Currently, one of our major research efforts focuses on the design of self-optimizing software systems. Real-world computer systems are never completely described by any theory. Because of this lack of theoretical understanding it is not possible to design high-performance algorithms that are oblivious to the network architecture and its underlying system software. Self-adaptive software attains high performance by allowing an application to adapt the computation to the significant details of the computing system, automatically. Successful experiences have shown the validity of using meta-programming for automatic performance optimization. The meta-program collects significant information on the performance of the application on a particular system and modifies it, accordingly. These experiences have been mainly carried out on shared memory parallel computers. We propose to extend this approach to local area networks (LANs) and the Grid.
W. Rivera, “Stability analysis of numerical boundary conditions in domain decomposition algorithms,” Journal of Applied Mathematics and Computation, vol. 137, 2003, pp. 375-385.
J. Seguel, D. Bollman, J. Feo ``A Framework for the Design and Implementation of FFT Permutation Algorithms''IEEE Transactions in Parallel and Distributed Systems, Vol. 11, No. 7, pp 625-635 (2000).
D. Bollman, J. Seguel, J. Feo - “A functional Approach to Radix-r FFTs”. Progress in Computer Research, Vol. I, Ed. F. Columbus, Nova Science Publishers, 2001, pp 77-103.
W. Rivera, "Numerical interface conditions for non-overlapping domain decomposition algorithms," Journal of Parallel and Distributed Computing. Submitted December 2001.
J. Seguel, D. Rodríguez; “The Doctoral Program in Computing and Information Sciences and Engineering of the University of Puerto Rico,” accepted for publication in Future Generation Computer Systems Special, issue on Computer Education, Elsevier Science B. V., Netherlands, 2003.
J. Seguel, D. Bollman, E. Orozco - New Prime Edge-Length Crystallographic FFT. To appear in Lecture Notes of Computer Sciences, ELSEVIER.
J. Seguel - A Unified Treatment of Symmetric FFT Code Generation. Conditionally accepted for publication in IEEE Transactions on Signal Processing.
J. Seguel and D. Burbano. "A Parallel Prime Edge-length Crystallographic FFT" to appear in Elsevier Lecture Notes of Computer Science.
J. Seguel, "Design and Implementation of Prime Edge-length Symmetric FFTs" to appear in Elsevier Lecture Notes of Computer Science.
J. Seguel, "Symmetric FFTs for Multidimensional Data Arrays of Prime Edge-length" conditionally accepted for publication in IEEE Transactions on Signal Processing.
W. Rivera, J. Zhu, and D. Huddleston, "An efficient parallel algorithm for solving unsteady nonlinear equations", Proceedings of the International Conference on Parallel Processing Workshops, Valencia, Spain, 2001, IEEE Computer Society, pp 79-84.
W. Rivera, "Adaptivity Support For Computational Grid-Aware Clusters," Proc. 6th World Multiconference on Systemics, Cybernetics and Informatics (SCI2002).
J. Yeckle and W. Rivera, “Mapping and characterization of applications in heterogeneous distributed systems,” To appear in Proc. 7th World Multiconference on Systemics, Cybernetics and Informatics (SCI2003).
F. Perez and W.Rivera, “An object-oriented framework for computational fluid dynamic simulations,” To appear in Proc. International Conference on Parallel and Distributed Processing Technology and Applications, (PDPTA2003).
W. Rivera, “CompuGe: A computational geometry learning environment.” To appear in Proc. IASTED International Conference on Computers and Advanced Technology in Education (CATE2003).
Book Chapters/Articles in Collections
W. Rivera, J. Zhu, and D. Huddleston, “An efficient parallel algorithm for solving unsteady Euler equations,” Parallel Computational Fluid Dynamics: Practice and Theory, P. Wilders, A. Ecer, J. Periaux, N. Satofuka, and P. Fox, eds., Elsevier Science, Amsterdam, 2002, pp. 293 – 300.