Partitioning around roadblocks: tackling constraints with intermediate relaxations
ICCAD '97 Proceedings of the 1997 IEEE/ACM international conference on Computer-aided design
An Introduction to Genetic Algorithms for Scientists and Engineers
An Introduction to Genetic Algorithms for Scientists and Engineers
Scheduling Divisible Loads in Parallel and Distributed Systems
Scheduling Divisible Loads in Parallel and Distributed Systems
A linear-time heuristic for improving network partitions
DAC '82 Proceedings of the 19th Design Automation Conference
Parallel Implementation of FDTD Computations Based on Macro Data Flow Paradigm
PARELEC '04 Proceedings of the international conference on Parallel Computing in Electrical Engineering
A parallel genetic algorithm based on global program state monitoring
PPAM'11 Proceedings of the 9th international conference on Parallel Processing and Applied Mathematics - Volume Part I
Hi-index | 0.00 |
In this paper, we discuss optimization of numerical computations of the FDTD problem in multiprocessor environments. The use of a genetic algorithm to find the best program macro data flow graph (MDFG) partition for a given FDTD problem for execution by a set of processors is presented. Different sub-graph merging actions are successively used in each step of the merging algorithm which starts from a program data flow graph representation. A special kind of chromosome represents consecutive steps of the graph partitioning algorithm to be applied to the current version of the macro data flow graph. To compare quality of individuals, we estimate the total execution time for each output MDF graph after applications of the actions specified in the algorithm, which they represent. To estimate efficiency of computations we used an architectural model which enables to represent parallel computations with 3 different communication protocols (MPI, RDMA RB, SHMEM). Experimental results obtained by simulation are presented.