M. Pelc (Poland)
parallel computing, computer control systems
In general, control algorithms with exact state observers possess rather sequential nature, because each control cycle consists of dependent components such as system input and output measurement, then state computation, then computation of control signal basing on system state and eventually application of the computed control signal to the system. The only possibility of improvement of the control algorithms performance as whole is parallelization of these components. In this paper the analysis of continuous state observers in the context of time expenses required for the observer computation is presented. The aim of this article is discussion especially, of what profits could be achieved by parallelizing of the computational process. The attention is mostly paid on checking how the computation task rank and the number of cluster nodes that perform computations influence on the control system performance. All of numerical tests were conducted by the use of Beowulf cluster and LAM/MPI library.. Detailed analysis of control algorithm with exact state observers resulted in the idea of implementation of double-layer decomposition approach to the control algorithm parallelization. Such an approach requires advanced or even expert knowledge about all of the control algorithm components, in return for it can give substantial improvement of the algorithm performance on the one side, and on the other side makes possible to choose the complexity of the control algorithm parallelization (the complexity depends on the level of decomposition).
Important Links:
Go Back
