Quality of service provisioning in wireless and high-speed networks
Upamanyu Madhow, H. Chaskar · 1999
The provision of quality of service (QoS) guarantees is a requirement in modern communication networks. The goal of this thesis is to investigate QoS provisioning in two technologies of current interest: wireless networking and high-speed wireline networking. High bit-error rates and time-varying channel quality are intrinsic to wireless links. The first part of this thesis focuses on supporting QoS on wireless links, the approach being to employ “link shaping” mechanisms that transform an unreliable wireless link into a near-lossless packet pipe in a manner tailored to the wireless channel characteristics, the traffic characteristics, and the QoS requirements of the application. These ideas are illustrated for Rayleigh faded wireless links. The first application considered is data communication, supported on TCP (the Internet transport protocol), over wireless. This thesis provides an approximate analysis, validated by computer simulations, for TCP performance over wireless links. The numerical results presented here show that a simple solution, that of using an appropriately designed retransmission-based link layer error recovery, prevents excessive deterioration of TCP throughput observed on raw wireless links. The second application considered is real time traffic seeking delay and loss guarantees over a wireless link. A framework for providing QoS is developed for a wireless downlink. The link shaping employs forward error correction with interleaving, and is combined with statistical multiplexing to obtain capacity gains. The goal in the second part of the thesis is the alleviation of a potential processing bottleneck encountered in packet scheduling in high-speed wireline networks. The scheduler should be designed to provide minimum-bandwidth and delay guarantees, and fair sharing of excess bandwidth. In the state-of-the-art schedulers, based on the weighted fair queuing (WFQ) paradigm, packets are assigned (transmission sequence) tags based on the bandwidth shares of the connections. In order to relieve the processing burden due to tag computation and sorting, variants of weighted round robin (WRR) scheduling paradigm are proposed in this thesis. Contrary to conventional thinking, it is shown that, for networks with fixed packet lengths, such WRR schemes can guarantee the same desirable properties as different WFQ schemes.