Web Delivery Performance Enhancement

As the widely use of web service, the performance of web delivery is critical in sustaining the web sites'popularity. The performance issues include the latency incurred by the end users in retrieving responses from web servers, the throughput of the server and so on. This thesis studied the web performance enhancement in wire and wireless network, proposed new admission control and scheduling algorithms to provide differentiated QoS for different classes. In conclusion, the contribution comprises of the following components:1.A new algorithm to provide relative and absolute delay guarantee for differentiated classes is proposed. We proposed a feedback driven approach-fuzzy controller which is suitable for the web server which is nonlinear and has no accurate model. The experiment results show that cmparing with the traditional PI controller, the fuzzy controller behaves low oscillation and can also adapt to various systems with different performance parameters.2.We studied how to guarantee another performance metric- percentile delay which the operators are more interested in. The percentile delay is gotten by ranking all delays in a sampling period, which brings much more nonlinearity to the system. We used the fuzzy controller to assign processes and made use of the additive-increase/multiplicative-decrease token shaper to do admission control for new arrived requests.3.Web delivery performance in wireless network is another hot research topic. We proposed a new PI controller to guarantee the handoff blocking probability to the predefined value (1%) under changing traffic when the system has no web traffic. When web traffic exists, we proposed an adaptive guard channel reservation algorithm to reserve reasonable resource to high prioirty classes to guarantee the blocking probability ratios among different classes. Simulation verified the proposed algorithms'efficiency.