Data structures and network algorithms
Data structures and network algorithms
Communications of the ACM
Hi-index | 0.00 |
The admittance method is an accurate approach for the analysis of electromagnetic circuits. Unfortunately, till now it has suffered from two main limitations: its high numerical complexity, and its lack of robustness, due to the risk of numerical ill-conditioning in a linear system representing the core of the approach.In this paper both drawbacks are solved, using a strategy based on the system partitioning into many independent and well-conditioned reduced-size subsystems, as well as on matrix thresholding. This approach, based on the exploitation of the matrix adjacence graph properties, allows a natural and straightforward migration on distributed platforms. The use of optimization strategies allows an optimum partitioning of the system in order to maximize the parallel efficiency.A master-slave and an SPMD implementation are proposed and compared on an eight-node IBM SP2.Results demonstrate the effectiveness of the serial strategy (ill-conditioning is avoided), and its efficiency in a parallel environment. Speed-ups quite near to Amdhal's limit are achieved by using the SPMD paradigm.