In: Proc. 16th Performance Engineering Workshop, 24-25 July 2000, Durham, UK, pages 165-176. 2000.
Abstract: A new parameter of quantitative fainess, or statistical fairness, describing behavior of asynchronous arbiters is proposed. It takes into account timing parameters of the arbiter and its environment described as stochastic process. In this context, a traditional qualitative fairness can be understood as a condition of having the probability of request serving equal to one under the exponential delay distribution. Thus the new parameter can be treated as a generalization of the common approach. The quantitative fairness of an ordered arbiter, which stores requests in a FIFO queue in the order of their arrival, has been studied using stochastic Petri net technique. An experiment has been conducted using the GreatSPN tool, whose results show a significant fairness improvement of an ordered arbiter in comparison with the corresponding arbiter without a FIFO. The method of quantitative fairness measurement in the GreatSPN environment has been developed.
Keywords: ordered arbiters, statistical fairness, stochastic Petri nets.