An analysis on error servers for stochastic network calculus

Network calculus is a recently developed theory dealing with queuing systems found in computer networks with focus on service guarantee analysis. In the current network calculus literature, the behavior of a server is typically modeled with the cumulative amount of service it successfully delivers, and the successfulness of service delivery implies no error in the delivered service. However, there are many networks such as wireless networks, where, not only is the service error-prone due to multi-access contention and/or random error on the communication link, but different error handling methods may also be applied. In such cases, it is difficult to directly apply the existing network calculus results due to lack of server models taking error into account. In this paper, an error server model is proposed for stochastic network calculus, based on which, an analysis on error servers is performed. The corresponding concatenation property is derived, which shows that under some general conditions, the tandem of error servers can be treated as an equivalent error server. In addition, to demonstrate the use and implication of the proposed error server model, performance bounds are derived and compared for a simple network. The study of the simple network shows that error handling may have significant impact on the performance bounds, and the proposed error server model can facilitate the analysis.