Delay analysis of a probabilistic priority discipline

Abstract

Abstract. In computer networks, the Strict Priority (SP) discipline is perhaps the most common and simplest method to schedule packets from different classes of applications, each with diverse performance requirements. With this discipline, however, packets at higher priority levels can starve packets at lower priority levels. To resolve this starvation problem, we propose to assign a parameter to each priority queue in the SP discipline. The assigned parameter determines the probability or extent by which its corresponding queue is served when the queue is polled by the server. We thus form a new packet service discipline, referred to as the Probabilistic Priority (PP) discipline. By properly adjusting the assigned parameters, not only is the performance of higher priority classes satisfied, but also the performance of lower priority classes can be improved. This paper analyzes the delay performance of the PP discipline. A decomposition approach is proposed for calculating the average waiting times and their bounds are studied. Two approximation approaches are proposed to estimate the waiting times. Simulation results that validate the numerical analysis are presented and examined. A numerical example which demonstrates the use of the PP discipline to achieve service differentiation is presented. This example also shows how the assigned parameters can be determined from the results of analysis mentioned above.

Achieving differentiated services through multi-class probabilistic priority scheduling

“…This paper extends the average delay analysis of the PP scheduling discipline in [10] to the multi-class case, derives several useful relationships concerning the average queueing delays of different classes and verifies the analytical results through simulation and implementation on a real DiffServ testbed. The implementation on the DiffServ testbed provides a proof of concept of the applicability of the PP discipline in real networks.…”

“…An in-depth analysis of the delay performance of the twoclass PP system is given in [10]. A decomposition approach was proposed for calculating the average queueing delay and upper and lower bounds were obtained.…”

“…We have analyzed the properties of the PP discipline and conducted extensive simulation studies of its performance [1], [10], [11], [12], which include comparisons with well-known scheduling disciplines such as Weighted Fair Queueing (WFQ) [13] and Weighted Round Robin (WRR) [14]. The PP discipline is considerably simpler to implement compared to WFQ since timestamping of packets is not required, there is no virtual and finish time computation and it does not require a sorted priority queue.…”

“…This section also includes a description of the formal and experimental results obtained so far for the PP discipline. Section III summarizes the results of a two-class PP system obtained in previous work [10]. In Section IV, we analyze the relationship between the average queueing delay of a particular class in a multi-class system and the assigned p i parameters.…”

Efficient implementation of relative differentiated services using proportional probabilistic priority

“…The authors in [5,6,9] showed that this algorithm exhibits the following properties that are very desirable to achieve service differentiation in a multi-class network by (a) providing diverse delay differentiation between classes, (b) supporting for weighted max-min fairness among classes, (c) overcoming the starvation problem inherent in SP, (d) supporting relative differentiated services, and (e) providing explicit bandwidth reservation guarantees by appropriate setting of a parameter…”