| Summary: | 博士 === 國立中興大學 === 應用數學系 === 90 === In order to support the requirements for the real-time transmission of multimedia application system, such as video conference, video on demand, voice over IP and so on, multi-service packet-switching networks must provide service guarantees to the connections, including guarantees on network delay, throughput, network delay-jitter, loss, and fairness. Anyhow, one of the most important issues in providing performance guaranteed service is the design of the packet-scheduling algorithm at each switching node. In packet-switched networks, packets come from different connections interact with each other at switching node. The packet-scheduling algorithms control the sequence in which packets are serviced and determine the relation of interaction at each connection.
In addition, for the most demanding applications, the network must offer a service, which can provide deterministic guarantees for the maximum delay of packets from all connections, referred as bounded delay service. To provide guaranteed service, resources have to be reserved within the network switching nodes. A typical user is concerned only with the quality of service of a network on an end-to-end basis. Therefore, how the end-to-end requirements are mapped into the local switching nodes requirement and the maximum network utilization is a critical function of the network internal design.
In the first part of this thesis, we propose a general framework based on the mapping of the end-to-end quality of service requirement into the local switching node quality of service requirements. We use the delay allocation policies under fair queuing scheduling algorithm at each switching node. With Leaky-bucket, (s,r), as the input traffic model, we derived a formula for delay bound and the number of connections. At the same time, we found that the delay bound as the quality of service metric; there is a significant difference in the performance of allocation policies. We also reference an evaluation strategy to analyze equal allocation policy and remaining capacity allocation policy. The numerical results compare these two allocation policies and show the correctness and efficiency.
In the second part of this thesis, we refer the result of first part to induce a general model for rate-based scheduling algorithms. This model can be applied for the rate-based packet scheduling algorithms, including Packet-by-packet Generalized Processor Sharing (PGPS), Self-Clocked Fair Queuing (SCFQ), and Starting-time Fair Queuing (SFQ). We apply two allocation policies, equal allocation and remaining capacity allocation to construct this model, and derive the relation between the network utilization and the allocation policy through mathematical analysis.
|
|---|
