Web cluster systems provide cost-effective solutions when scalable and reliable web services are required. However, as the number of servers in web cluster systems increase, web cluster systems incur long and unpredictable delays to manage servers. This study presents the efficient management schemes for web cluster systems. First of all, we propose an efficient request distribution scheme in web cluster systems. Distributor-based systems forward user requests to a balanced set of waiting servers in complete transparency to the users. The policy employed in forwarding requests from the frontend distributor to the backend servers plays an important role in the overall system performance. In this study, we present a proactive request distribution (ProRD) to provide an intelligent distribution at the distributor. Second, we propose the heuristic memory management schemes through a web prefetching scheme. For this study, we design a Double Prediction-by-Partial-Match Scheme (DPS) that can be adapted to the modern web frameworks. In addition, we present an Adaptive Rate Controller (ARC) to determine the prefetch rate depending on the memory status dynamically. For evaluating the prefetch gain in a server node, we implement an Apache module. Lastly, we design an adaptive web streaming system in wireless networks. The rapid growth of new wireless and mobile devices accessing the internet has contributed to a whole new level of heterogeneity in web streaming systems. Particularly, in-home networks have also increased in heterogeneity by using various devices such as laptops, cell phone and PDAs. In our study, a set-top box(STB) is the access pointer between the internet and a home network. We design an ActiveSTB which has a capability of buffering and quality adaptation based on the estimation for the available bandwidth in the wireless LAN.

The demand for ever larger, more efficient, reliable and cost effective communication networks necessitates new network architectures, such as wireless ad hoc networks, cognitive radio, relaying networks, and wireless sensor networks. The study of such networks requires a fundamental shift from thinking of a network as a collection of independent communication pipes, to a multi-user channel where users cooperate via conferencing, relaying, and joint source-channel coding. The traditional centralized networks, such as cellular networks, include a central controller and a fixed infrastructure, in which every node communicates with each other via a centralized based station (BS). However, for a decentralized network, such as wireless ad hoc networks and wireless sensor networks, there is no infrastructure support and no central controllers. In such multi-user wireless networks, the scheduling algorithm plays an essential role in efficiently assigning channel resources to different users for better system performance, in terms of system throughput, packet-delay, stability and fairness. In this dissertation, our main goal is to develop practical scheduling algorithms in wireless ad hoc networks to enhance system performance, in terms of throughput, delay and stability. Our dissertation mainly consists of three main parts. First, we identify major challenges intrinsic to ad hoc networks that affect the system performance, in terms of throughput limits, delay and stability condition. Second, we develop scheduling algorithms for wireless ad hoc networks, with various considerations of non-cooperative relays and cooperative relays, fixed-rate transmission and adaptive-rate transmission, full-buffer traffic model and finite-buffer traffic model. Specifically, we propose an opportunistic scheduling scheme and study the throughput and delay performance, with fixed-rate transmissions in a two-hop wireless ad hoc networks. In the proposed scheduling scheme, we prove two key inequalities that capture the various tradeoffs inherent in the broad class of opportunistic relaying protocols, illustrating that no scheduling and routing algorithm can simultaneously yield lower delay and higher throughput. We then develop an adaptive rate transmission scheme with opportunistic scheduling, with the constraints of practical assumptions on channel state information (CSI) and limited feedback, which achieves an optimal system throughput scaling order. Along this work with the consideration of finite-buffer model, we propose a Buffer-Aware Adaptive (BAA) scheduler which considers both channel state and buffer conditions to make scheduling decisions, to reduce average packet delay, while maintaining the queue stability condition of the networks. The proposed algorithm is an improvement over existing algorithms with adaptability and bounded potential throughput reduction. In the third part, we extend the methods and analyses developed for wireless ad hoc networks to a practical Aeronautical Communication Networks (ACN) and present the system performance of such networks. We use our previously proposed scheduling schemes and analytical methods from the second part to investigate the issues about connectivity, throughput and delay in ACN, for both single-hop and two-hop communication models. We conclude that the two-hop model achieves greater throughput than the single-hop model for ACN. Both throughput and delay performances are characterized.

No description available.