QUEUE LENGTH ANALYSIS OF END-TO-END DIFFERENTIATED SERVICE NETWORKS WITH SELF-SIMILAR TRAFFIC By

Name: YUE WANGDate of Degree: July, 2005Institution: Oklahoma State UniversityLocation: Stillwater, OklahomaTitle of Study: QUEUE LENGTH ANALYSIS OF END-TO-END DIFFERENTIATEDSERVICE NETWORKS WITH SELF-SIMILAR TRAFFICPages in Study: 74Candidate for the Degree of Master of ScienceMajor Field: Computer ScienceScope and Method of Study: This thesis focuses on the analysis of queue lengthboundaries in end-to-end self-similar networks with differentiated service. First,the Hurst parameter was calculated by three methods. Secondly, queue lengthboundaries were estimated at a single hop with differentiated service under self-similar traffic. Finally, the derivations are extended to end-to-end differentiatedservice networks with self-similar traffic.Findings and Conclusions: Quality-of-Service (QoS) is a key issue in networks.Improving the performance with guaranteed QoS is one of the major problems. Itis well known that traditional analytical methods of queuing systems are based onPoisson and Poisson-based stochastic processes. Unfortunately, these methods,like Jackson theorem, are not applicable in high speed broadband networks. Inthis thesis, a novel analytical model is proposed based on the arrival rate and theservice rate for multiple hops queuing systems. Then the mathematical derivationsare extended to end-to-end differentiated service networks with self-similartraffic. The upper and lower bound of the queue length at each hop is derived. Theresults illustrate the performance gain in queue length at each hop. Finally, anovel adaptive admission controller algorithm is proposed based on the arrivalrate and the service rate for multiple hop queuing systems with self-similarnetwork traffic. The adaptive admission controller algorithm can provide aguaranteed QoS performance for the higher priority classes, as long as the highestclass QoS requests do not exceed the comprehensive network resources. ADVISER’S APPROVAL: Dr. G. E. Hedrick