Traffic self-similarity (i.e., noticeable burstiness over multiple time scales) has been discovered to be a ubiquitous phenomenon in modern communication networks and multimedia systems. Self-similar traffic can considerably deteriorate user-perceived Quality- of-Service (QoS) and has drawn significant interest and received tremendous research efforts from both academia and industry. Due to its fractal-like nature, performance modelling of self-similar traffic poses greater challenges and exhibits more complexity than that of traditional non-bursty traffic.
The hybrid scheduling scheme that integrates the well-known Priority Queueing (PQ) and Generalized Processor Sharing (GPS) has been suggested as a promising mechanism for the provisioning and implementation of differentiated QoS. In this talk, we will present a new analytical model we have recently developed for the hybrid PQ-GPS scheduling scheme subject to self-similar traffic.
In order to make the challenging performance modelling problem tractable, we propose a novel flow-decomposition approach that can equivalently divide the PQ-GPS system into a group of Single-Server Single-Queue (SSSQ) systems and derive their service capacities. Based on the Large Deviation Principle and by virtue of the equivalent relationship between these SSSQ systems and the original PQ-GPS system, we derive the queue length distributions and loss probabilities of individual traffic flows in the hybrid system. The validity of the proposed flow-decomposition approach and the accuracy of the analytical model are demonstrated by comparing analytical results to those obtained through extensive simulation experiments of theactual system.