The Internet is an essential resource to an ever-increasing number of businesses and home users. Internet access is increasing dramatically and hence, the need for efficient and effective Web server systems is on the rise. These systems are information engines that are accessed through the Internet by a rapidly growing client base. These systems are expected to provide good performance and high availability to the end user. They are also resilient to failures at both the hardware and software levels. These characteristics make them suitable for servicing the present and future information demands of the end consumer. In recent years, researchers have concentrated on taxonomies of scalable Web server system architectures, and routing and dispatching algorithms for request distribution. However, they have not focused on the classification of commercial products and prototypes, which would be of use to business professionals and software architects. Such a classification would help in selecting appropriate products from the market, based on product characteristics, and designing new products with different combinations of server architectures and dispatching algorithms. Currently, dispatching algorithms are classified as content-blind, content-aware, and Domain Name Server (DNS) scheduling. These classifications are extended, and organized under one tree structure in this thesis. With the help of this extension, this thesis develops a unified product-based taxonomy that identifies product capabilities by relating them to a classification of scalable Web server systems and to the extended taxonomy of dispatching algorithms. As part of a detailed analysis of Web server systems, generic queuing models, which consist of a dispatcher unit and a Web server unit are built. Some performance metrics, such as throughput, server performance, mean queue size, mean waiting time, mean service time and mean response time of these generic queuing models are measured for evaluation. Finally, the correctness of generic queuing models are evaluated with the help of theoretical and simulation analysis. / May 2005

The Internet is an essential resource to an ever-increasing number of businesses and home users. Internet access is increasing dramatically and hence, the need for efficient and effective Web server systems is on the rise. These systems are information engines that are accessed through the Internet by a rapidly growing client base. These systems are expected to provide good performance and high availability to the end user. They are also resilient to failures at both the hardware and software levels. These characteristics make them suitable for servicing the present and future information demands of the end consumer. In recent years, researchers have concentrated on taxonomies of scalable Web server system architectures, and routing and dispatching algorithms for request distribution. However, they have not focused on the classification of commercial products and prototypes, which would be of use to business professionals and software architects. Such a classification would help in selecting appropriate products from the market, based on product characteristics, and designing new products with different combinations of server architectures and dispatching algorithms. Currently, dispatching algorithms are classified as content-blind, content-aware, and Domain Name Server (DNS) scheduling. These classifications are extended, and organized under one tree structure in this thesis. With the help of this extension, this thesis develops a unified product-based taxonomy that identifies product capabilities by relating them to a classification of scalable Web server systems and to the extended taxonomy of dispatching algorithms. As part of a detailed analysis of Web server systems, generic queuing models, which consist of a dispatcher unit and a Web server unit are built. Some performance metrics, such as throughput, server performance, mean queue size, mean waiting time, mean service time and mean response time of these generic queuing models are measured for evaluation. Finally, the correctness of generic queuing models are evaluated with the help of theoretical and simulation analysis.

The Internet is an essential resource to an ever-increasing number of businesses and home users. Internet access is increasing dramatically and hence, the need for efficient and effective Web server systems is on the rise. These systems are information engines that are accessed through the Internet by a rapidly growing client base. These systems are expected to provide good performance and high availability to the end user. They are also resilient to failures at both the hardware and software levels. These characteristics make them suitable for servicing the present and future information demands of the end consumer. In recent years, researchers have concentrated on taxonomies of scalable Web server system architectures, and routing and dispatching algorithms for request distribution. However, they have not focused on the classification of commercial products and prototypes, which would be of use to business professionals and software architects. Such a classification would help in selecting appropriate products from the market, based on product characteristics, and designing new products with different combinations of server architectures and dispatching algorithms. Currently, dispatching algorithms are classified as content-blind, content-aware, and Domain Name Server (DNS) scheduling. These classifications are extended, and organized under one tree structure in this thesis. With the help of this extension, this thesis develops a unified product-based taxonomy that identifies product capabilities by relating them to a classification of scalable Web server systems and to the extended taxonomy of dispatching algorithms. As part of a detailed analysis of Web server systems, generic queuing models, which consist of a dispatcher unit and a Web server unit are built. Some performance metrics, such as throughput, server performance, mean queue size, mean waiting time, mean service time and mean response time of these generic queuing models are measured for evaluation. Finally, the correctness of generic queuing models are evaluated with the help of theoretical and simulation analysis.

Server farms are popular architectures for computing infrastructures such as supercomputing centers, data centers and web server farms. As server farms become larger and their workloads more complex, designing efficient policies for managing the resources in server farms via trial-and error becomes intractable. In this thesis, we employ stochastic modeling and analysis techniques to understand the performance of such complex systems and to guide design of policies to optimize the performance. There is a rich literature on applying stochastic modeling to diverse application areas such as telecommunication networks, inventory management, production systems, and call centers, but there are numerous disconnects between the workloads and architectures of these traditional applications of stochastic modeling and how compute server farms operate, necessitating new analytical tools. To cite a few: (i) Unlike call durations, supercomputing jobs and file sizes have high variance in service requirements and this critically affects the optimality and performance of scheduling policies. (ii) Most existing analysis of server farms focuses on the First-Come- First-Served (FCFS) scheduling discipline, while time sharing servers (e.g., web and database servers) are better modeled by the Processor- Sharing (PS) scheduling discipline. (in) Time sharing systems typically exhibit thrashing (resource contention) which limits the achievable concurrency level, but traditional models of time sharing systems ignore this fundamental phenomenon. (iv) Recently, minimizing energy consumption has become an important metric in managing server farms. State-of-the-art servers come with multiple knobs to control energy consumption, but traditional queueing models don’t take the metric of energy consumption into account. In this thesis we attempt to bridge some of these disconnects by bringing the stochastic modeling and analysis literature closer to the realities of today’s compute server farms. We introduce new queueing models for computing server farms, develop new stochastic analysis techniques to evaluate and understand these queueing models, and use the analysis to propose resource management algorithms to optimize their performance.

Les serveurs informatiques permettent de mettre en œuvre un large éventail de services en ligne et d'applications telles que les services Web, les services de messagerie ou les services de base de données. Cependant, leur configuration ad hoc soulève des problématiques cruciales liées à la performance, la disponibilité et le coût économique des applications. Dans cette thèse, nous étudions l'impact de la configuration des serveurs sur le compromis central entre performance et disponibilité du service. Dans une première partie, nous modélisons le comportement de serveurs uniques à l'aide d'approximations fluides et développons des lois de commandes novatrices pour le contrôle d'admission sur les systèmes de serveurs centralisés. Nous proposons plusieurs lois de commande pour différentes combinaisons d'objectifs de qualité de service et de niveau de service. Parmi elles, AM-C garantit un objectif de performance du service en maximisant sa disponibilité; PM-C garantit un objectif de disponibilité du service en maximisant sa performance. Nous évaluons expérimentalement le modèle ainsi que les stratégies de contrôle sur TPC-C, un benchmark industriel reconnu qui reproduit un service d'e-commerce, implémenté sur le serveur de base de données PostgreSQL. Nos expériences montrent que les techniques proposées améliorent jusqu'à 30% les performances du service tout en garantissant les contraintes de disponibilité. Dans une deuxième partie, nous étendons ce travail aux systèmes de serveurs distribués, qui sont largement utilisés par les applications Internet évoluant sur des systèmes multi-niveaux hébergés par des clusters de serveurs. Nous présentons un modèle de serveur distribué comme un modèle continu non linéaire en faisant une analogie avec les systèmes transferts de fluide. Nous formalisons alors un problème d'optimisation pour le contrôle de ces systèmes. Nous fournissons un contrôle d'admission qui permet d'obtenir la disponibilité de service la plus élevée tout en garantissant un objectif de performance. Une évaluation numérique du modèle proposé et du contrôle distribué associé est présentée et montre que la configuration optimale de ces systèmes n'est pas intuitive. / Server technology provides a means to support a wide range of on-line services and applications, such as web services, e-mail services, database services. However, their ad hoc configuration poses significant challenges to the performance, availability and economical costs of applications. In this thesis, we examine the impact of server configuration on the central trade-off between service performance and service availability. First, we model the behavior of single servers using fluid approximations. Second, we develop novel ad- mission control laws of central server systems. We provide several control laws for different combinations of quality-of-service and service level objectives. Among them, AM-C , the availability-maximizing admission control law, achieves the highest service availability while meeting given performance objective; PM-C is a performance-maximizing ad- mission control law that meets a desired availability target with the highest performance. We evaluate our fluid model and control techniques on the TPC-C industry-standard benchmark that implements a warehouse running on the PostgreSQL database server. Our experiments show that the proposed techniques successfully improve performance by up to 30 % while guaranteeing availability constraints. Furthermore, we extend this work to distributed server systems, that are widely used by Internet applications in the farm of server clusters and multi-tier systems. We present a distributed server model as a non-linear continuous-time model using analogies with fluid transfer. We then state an optimization problem for the control of distributed server systems. We provide an admission control that allows to get the highest service availability while a target performance level is guaranteed. Numerical evaluations of the proposed distributed model and control are presented, and show that the optimal configuration of such systems is not intuitive.

Although several studies question information technology's contribution to productivity, organizations continue to invest in client/server systems (CSSs) particularly as enablers of business process reengineering (BPR). These efforts may be wasted if they do not improve business processes. This study focused on business processes and investigated the role of workstation-based CSSs in changing business processes. A multiple case study of workstation-based CSS databases in three organizations was performed with the proposition that they moderate the relation between managerial action and changes within business processes. The research framework suggested that changes to business processes are achieved by reducing uncertainty. In order to measure change in business processes, this study categorized business process change into: (1) compressing sequential tasks across functions, (2) compressing tasks vertically within the managerial hierarchy, (3) eliminating slack resources, (4) reducing the distance between the point of decision and the point of information or eliminating intermediaries, (5) reconfiguring sequential processes to operate in parallel, and (6) linking parallel activities during the process. Data collected from questionnaires, interviews, and observations from three case studies were used to construct network diagrams, relationship matrices, reachability matrices, and task tables of business processes. The results of this research partially support the proposition that managerial action affects business process change by reducing uncertainty. This research suggests that changes in the use of workstation-based CSSs are related to changes in business processes. However, because ofthe small sample size, no finding was made regarding changes in the strength of that relationship. Therefore, within its limitations, this research (1) partially supports the proposition that CSSs moderate changes in business processes, (2) found that both favorable and unfavorable changes may result from using CSSs, (3) explains how business process change occurs, and (4) suggests new variables for measuring successful BPR.