Relative Service Differentiation over Bandwidth on Demand Satellite Networks
Authors: Chai, W.K.
Editors: Pavlou, G.
A key to next generation Internet is service differentiation. Broadband satellite networks being an integral and complementary part of this global information infrastructure should therefore be capable of providing differential treatment to users with different quality of service (QoS) requirements. This thesis addresses the problem of providing relative service differentiation over next generation satellite networks. We develop a complete service differentiation framework for broadband geostationary (GEO) bandwidth on demand satellite networks that provides proportional differentiated services (PDS) to different network performance metrics. The framework offers a separate "tuning knob" for satellite operators to control the differentiation of each performance metric.
We consider the differentiation of three most important QoS metrics: packet queuing delay, transmission control protocol (TCP) throughput and packet loss. We begin with proportional delay differentiation and propose a time dependent scheduler operating at the medium access control (MAC) layer, named Satellite Waiting Time Priority scheduler. The scheduler closely emulates the PDS model even in short timescales. However, the feasibility region of the scheduler is limited to high load conditions.
We focus our throughput differentiation investigation on TCP traffic as it forms the majority of Internet data traffic. Employing the cross-layer approach, we design an integrated solution that requires the joint configuration of TCP-Performance Enhancing Proxy at the transport layer and the scheduling algorithm controlling the resource allocation at the MAC layer. The complementary behaviours of the two differentiation mechanisms combine to achieve the desired differentiation.
Realising the need to automate the process of configuring the differentiation parameters, we further explore the use of fixed-point methods to analytically approximate the obtained differentiated TCP performance over dynamic load conditions. We validate our analytical method with extensive simulation results showing close agreement between the approximated and simulated differentiation of TCP flows.
For loss differentiation, we propose an innovative profile-based probabilistic dropping scheme. It is capable of providing consistent network-wide proportional loss rate to traffic of different service classes. Although capable of achieving the PDS model under different load conditions, we find instances where the predictability property of the PDS model is violated under certain configurations.
Finally, we depart from the PDS model and examine the congestion pricing approach in realising relative service differentiation over both GEO and non-GEO satellite networks. We implement and examine a window-based congestion pricing algorithm that differentiates users based on their respective willingness to pay to achieve the most basic requirement of relative service differentiation i.e. high priority traffic always receive better or no worse performance compared to low priority traffic. We finally conclude that although such an approach may be technically feasible for satellite networks, its adoption in the real world may depend on a lot of other factors.