Optimised queuing strategies for multi-level QoS

Abstract

A variety of network applications, such as FTP, HTTP, video on-demand, IP phone, have different requirements on traffic parameters, such as bandwidth, loss rate, delay bound and delay jitter. In order to provide the guaranteed service on above four traffic parameters, the concept of Quality of Service (QoS) and traffic management is essential to apply on the end-to-end paths of network applications. Most QoS scheduling algorithms applies a single-parameter approach; they are based on either bandwidth or delay. In order to support more than one QoS parameter, the decoupling of bandwidth and delay in queuing can be deployed and this has been proved a difficult problem. Although recent research has tried to use rate regulator and priority-based scheduler to guarantee bandwidth and delay criteria respectively, the outgoing performance is insufficiently acceptable in order to be implemented in the present network environments. In this research, a new scheduling algorithm called Delay-Differentiable Fair Queuing (D2FQ) algorithm is proposed. This algorithm is a two-parameter approach that decouples the bandwidth and delay into a single model. Although D2FQ consists of the functions of rate regulator and delay scheduler, it has a low time complexity, O(log(L)), where L is a key component in D2FQ scheduler. This scheduling algorithm is also practicable for industrial applications. This thesis will also show the fairness analysis of D2FQ where the fairness on the bandwidth allocation is the basic criterion for QoS. The computation complexity and the delay bound will also be analyzed in detail. For the purpose of simulation discussed in this thesis, a single-node network topology and multi-node network topology will be applied. Single-node network topology can simplify the network environment to facilitate the performance analysis of the scheduler in respect to fairness of bandwidth allocation. In this thesis, a single-node network will be applied on connectionless traffic analysis, which includes the throughput fairness and differentiate delay fairness on different offered loads and different traffic conditions. Some famous schedulers, such as WFQ, DRR and CSFQ, are used for performance comparison with D2FQ. The two traffic parameters of D2FQ, throughput weight and delay weight, produce a more complex traffic condition and provide a more useful analysis on D2FQ. A single-node network is generally used for connectionless traffic analysis. Most of the packet schedulers only use a single-node network topology. In addition to the propositions stated above, multi-node network topology will also be applied on connection-oriental traffic analysis in this thesis. Analysis on connection-oriented flow is useful because most of the traffics in present network are connection-oriented flow and the multi-node network topology can increase the increase the credibility of performance on D2FQ. The analysis for simulation on multi-node network topology includes drop rate and delay on the condition of overloading. The simulation results show that D2FQ has a good performance on throughput analysis and flow isolation, which conclude that D2FQ can fairly decouple throughput and delay.