A Mapping Strategy for Parallel Processing
IEEE Transactions on Computers
A Partitioning Strategy for Nonuniform Problems on Multiprocessors
IEEE Transactions on Computers
Solving problems on concurrent processors
Solving problems on concurrent processors
Nearest-neighbor mapping of finite element graphs onto processor meshes
IEEE Transactions on Computers
Heuristic approaches to task allocation for parallel computing
Heuristic approaches to task allocation for parallel computing
SIGARCH Third Conference on Hypercube Concurrent Computers and Applications
Introduction to algorithms
Partitioning sparse matrices with eigenvectors of graphs
SIAM Journal on Matrix Analysis and Applications
Performance of dynamic load balancing algorithms for unstructured mesh calculations
Concurrency: Practice and Experience
Physical optimization algorithms for mapping data to distributed-memory multiprocessors
Physical optimization algorithms for mapping data to distributed-memory multiprocessors
Numerical Algorithms for Modern Parallel Computer Architectures
Numerical Algorithms for Modern Parallel Computer Architectures
Multilevel circuit partitioning
DAC '97 Proceedings of the 34th annual Design Automation Conference
Congestion and almost invariant sets in dynamical systems
SNSC'01 Proceedings of the 2nd international conference on Symbolic and numerical scientific computation
| Hi-index | 0.00 |
Mapping data to parallel computers aims at minimizing the executiontime of the associated application. However, it can take anunacceptable amount of time in comparison with the execution timeof the application if the size of the problem is large. In thisarticle, first we motivate the case for graph contraction as ameans for reducing the problem size. We restrict our discussion toapplications where the problem domain can be described using agraph (e.g., computational fluid dynamics applications). Then wepresent a mapping-oriented parallel graph contraction (PGC)heuristic algorithm that yields a smaller representation of theproblem to which mapping is then applied. The mapping solution forthe original problem is obtained by a straightforwardinterpolation. We then present experimental results on usingcontracted graphs as inputs to two physical optimization methods;namely, genetic algorithm and simulated annealing. The experimentalresults show that the PGC algorithm still leads to a reasonablygood quality mapping solutions to the original problem, whileproducing a substantial reduction in mapping time. Finally, wediscuss the cost-quality tradeoffs in performing graph contraction.