Effect of service modeling on medium access control performance

Abstract

Due to the tractability in analytical methods exponential burst distributions are more common in service modeling literature. However, previous traffic measurements findings argue heavy tailed distributions such as Pareto represents burst sizes in WWW browsing and Cauchy distributed burst sizes in e-mail sessions. The effect of long range dependence (LRD) introduced by these heavy tailed distributions results in self similarity in the aggregate traffic. The impact of self similarity is significant in the area of queue delay performance characteristics. This work addresses the issues of service modeling in medium access control (MAC) performance compared to traditional exponential session distributions against the above mentioned distributions in WWW browsing and e-mail sessions applied to an example GPRS MAC system. The results conclude high delay in Cauchy e-mail connection sizes and Pareto WWW burst sessions compared to exponential connections/bursts. Also, once looking in to the matter it can be seen that this delay is mainly due to the congestion caused by different types of bursts/connection distributions. Therefore the bursty traffic such as WWW browsing which demonstrates self similarity at the aggregate level influence the MAC protocol performance greater than in traditional traffic models.