Performance analysis of degradable multiprocessor systems

A method for performance evaluation of degradable multiprocessor systems with random failure and repair using hierarchical generalized stochastic Petri net (GSPN) model is presented. The failure and repair rates in a multiprocessor system are an order of magnitude smaller than the rates of job arrival and completion; therefore, the GSPN model can be decomposed into a hierarchical sequence of aggregated subnets, each of which is valid at a certain time scale. These smaller subnets can be solved in isolation and their results combined to get the solution of the hole system. The model of the aggregated network has a product-form solution, so performance analysis in the presence of failure and repair can be carried out without the additional complexity of a fault-free model. A multiprocessor system is analyzed which is intractable using conventional solution techniques. The complexity of the time-scale decomposition technique is found to be same as fault-free model without any significant loss in the accuracy of the result.<<ETX>>