A framework for service differentiating Internet servers

The variances in application types and the service quality requirements have been increasing continuously. The current best-effort service model of the Internet and its servers might not be able to satisfy the evolving service quality demands of diverse applications. Therefore, diferentiated service has been proposed as a potential solution to provide alternative quality of services (QoS) and is expected to be supported in the next generation Internet (NGI). The Service Differentiating Internet Server (SDIS) is proposed in this dissertation to provide service guarantee to prioritized client requests, which enforces predictable QoS by categorizing requests into different groups based on task types and client; identifications and allocating resources based on the groups. Considering the popularity of the WWW on the Internet and its capability of providing a common interface to almost all the protocols on the Internet, Web server performance issues are emphasized in this study.; Four major steps are taken to conduct this study. First, the workload behavior of Web servers under different; environments are studied. The size, access pattern, lifetime distribution, and behavior of modification of different types of documents in different classes of web sites are analyzed and compared. Results indicate significant differences in static as well as dynamic characteristics of documents types in different web site classes, which suggests that server workload characterization based on the application environments helps to improve the web server caching efficiency and thus improve the server throughput.; Second, the conceptual correctness of service differentiating Internet servers (SDIS) is examined and verified. Approaches of providing differentiated and QoS based web services, and modeling and analysis of web server scheduling under different workload situations are studied. Experimental studies and analyses prove that a SDIS provides significantly better services to high priority tasks compared to a traditional Internet server under high system utilization.; Third, the overload management versus QoS assurance in a web server is explored. The performance of a overloaded server becomes unstable and may eventually crash. To assure stable and predictable service, a simple and efficient admission control algorithm ACES is proposed. The ACES algorithm bounds response delay of requests by allocation of computational credits based on the estimation of task service times. Experimental study shows that the ACES algorithm provides effective control of response delay bounds under highly variant workload environments while maintaining the system throughput.; We further extend the research on performance assurance issues of SDIS, and present a novel and efficient admission control algorithm, PACERS, which provides services based on the server workload characteristics. Different levels of quality of services are assured by periodic allocation of system resources based on the estimation of request rate and service requirements of prioritized tasks. Theoretical analysis and experimental study show that the PACERS algorithm provides effective control of throughput and response delay boundary to the prioritized tasks, and preserves system throughput under various workload situations.; In summary, this research has the following contributions: traffic characterization which is essential for designing efficient and high performance Internet servers; the conceptualization of SDIS framework and its design to meet the service quality demands of evolving applications and services; development of efficient admission and overload control techniques aiming at performance assurance of Internet servers.