A neural network approach to process assignment in multiprocessor systems based on the execution time

The distribution of tasks among the various processors available is a crucial problem in a distributed environment. We propose a new method for process assignment to processors by using the predicted time of execution as the parameter that decides the assignment. This execution time is predicted by means of a neural network. We have designed a new architecture with a self organizing map as a hidden nodes selector for a back propagation network in our design. The SOM takes the process characteristics as input and classifies the process. A back propagation network is used to predict the execution time taking the system characteristics as input and the nodes in the network are chosen by the winning neuron in the SOM classifier. The structure of the back propagation network is determined by which neuron wins in the SOM (process classifier). We have simulated this and have found that our model has been accurate in predicting execution time.