Modeling and performance evaluation of multiprocessor systems for real-time nonlinear robot control

The implementation of a nonlinear control algorithm in multiprocessor systems is addressed. A single bus structure and a multiport structure are proposed for real-time computation. The computing task is decomposed into a reasonable number of subtasks by considering both the execution and the communication time of subtasks. In order to deal with the real-time performance, the models of multiprocessor systems with single-bus and multiport-memory architectures are established using Markovian birth-death queuing models. The intercommunication time can thus be estimated using analytical formulas. On the basis of the estimated intercommunication time and task execution time, the necessary number of processors in the multiprocessor system for a specific real-time robot control algorithm can be estimated.<<ETX>>