Statistical traffic modeling and guaranteed service disciplines: a performance evaluation paradigm

In this paper, a new performance analysis framework for guaranteed services is introduced. Performance is evaluated in terms of the loss probability and delay distribution statistically guaranteed. Traffic sources are characterized statistically and are modeled with Markov-based processes. The service is guaranteed by allocating a service curve to each connection and some buffer space. As service curve based servers are not memoryless, the resulting queueing system, called SBBP/Sc/1/K, is not Markovian. For this reason, the system is observed only at some renewal instants. The process obtained is Markovian and can be easily solved. Then we use this solution to evaluate the steady-state probabilities of the SBBP/Sc/1/K process which are needed for the performance analysis. (C) 2001 Elsevier Science B.V. All rights reserved.