| The widespread growth of the World Wide Web made information easy to disseminate and distribute over a network of computer networks known as the Internet. This phenomenon also demonstrated that the time it takes a piece of content to reach the user is an important issue in measuring the performance of delivering information in an effective manner. On the path from the source of the content to the end user, several solutions to reduce user delay have been implemented. Among these, placing the content as close as possible to the end users who are at the edges of the computer networks has been widely used.;The goal of this research effort is to improve the techniques for reducing the delay users experience while browsing the web. We address the problem in the context of web proxy server caching and web content distribution.;We evaluate an admission control mechanism that augments the LRU (Least Recently Used) algorithm used in web proxy servers. The algorithm admits the documents based on their attributes and classifies them as cacheable or non-cacheable. The cacheable documents are kept in the proxy server and are served from there to the end user when requested. Our results are useful for Internet Service Providers (ISPs) or for an enterprise network which runs a forward proxy server for its employees and the results show substantial improvements in reducing user delay.;On the web content distribution side, we explore the viability of a content distribution network which provides dynamic routing and efficient distribution of content to reduce user delay. First we develop an operational and economic model of Content Distribution which evaluates two options a content provider (CP) may have for distributing its content: (1) forming arrangements with the ISPs by itself and deploying servers in their networks, and (2) by subscribing to the services of a Content Distribution Network Provider (CDP). In this context, we conclude that the CDN services make sense when the CPs are fragmented, i.e. no one CP dominates in terms of demand. CDN services are also beneficial to the CP when the users are also fragmented, in other words users are distributed evenly, and no one user location has a high demand concentration.;Next, we examine the capacity decisions of the CP if it decides to distribute its content by placing servers in different networks connected to the Internet. Of particular importance in this study is that each CP site will face surges in its traffic. In order to maintain the quality of service of its websites the CP must make intelligent decisions when planning its capacity, however may adjust its capacity to an approximate Poisson traffic, wrongly assuming the surges come from a Poisson distribution. This assumption may not capture the variability of the real traffic, hence, as the traffic burstiness increases, the CP will put additional capacity to meet the desired level of user delay also increases if it wants to avoid downtime. When one of the servers of the CP fails and its traffic is redirected, the capacity requirements increase when the original servers were operating on a relatively high utilization levels. In other cases, the Poisson approximation by the CP suffices to maintain the quality level of service, described by the total expected waiting time for the user. |
