Packet scheduling strategies for emerging service models in the internet

Hongyuan Shi

doi:10.17918/etd-217

Traditional as well as emerging new Internet applications such as video-conferencing and live multimedia broadcasts from Internet TV stations will rely on scheduling algorithms in switches and routers to meet a diversity of service requirements desired from the network. This dissertation focuses on four categories of service requirements that cover the vast majority of current as well as emerging new applications: best-effort service, guaranteed service (delay and bandwidth), controlled load service, and soft real-time service. For each of these service types, we develop novel packet scheduling strategies that achieve better performance and better fairness than existing strategies. Best-effort and guaranteed services: A fair packet scheduler designed for best- effort service can also be employed to achieve bandwidth and delay guarantees. This dissertation proposes a novel fair scheduling algorithm, called Greedy Fair Queueing (GrFQ),that explicitly incorporates the goal of achieving better fairness into the actions of the scheduler. A simplified version of the scheduler is also proposed for easier deployment in real networks. Controlled load service: This dissertation analyzes and defines requirements on packet schedulers serving traffic that request the controlled load service (part of the Integrated Services architecture). We then propose a novel scheduler, called the CL(®) scheduler, which provides service differentiation for aggregated traffic for controlled load service. The proposed scheduler satisfies the defined requirements with a very low processing complexity and without requiring per-flow management. Soft real-time service: We formally define the service requirements of soft real-time applications which have delay constraints but which can tolerate some packet losses. Two novel schedulers of different levels of complexity are proposed. These schedulers achieve better performance (lower overall loss rates) and better fairness than previously known schedulers. We adapt a metric used widely in economics, called the Gini index, to our purpose of evaluating the fairness achieved by our schedulers under real traffic conditions. The Gini index captures the instantaneous fairness achieved at most instants of time as opposed to previously used measures of fairness in the networking literature. Using real video, audio and gateway traffic traces, we show that the proposed schedulers achieve better performance and fairness characteristics than other known schedulers.

Packet scheduling strategies for emerging service models in the internet

Files and links (1)

Abstract

Metrics

Details

Packet scheduling strategies for emerging service models in the internet

Files and links (1)

Abstract

Metrics

Details

Drexel University Social media