E-business (electronic business) is the conduct of business electronically, including purchasing, selling, servicing customers, and collaborating with business partners. It streamlines supply chain, manufacturing, and procurement systems to deliver the right products and services to the right customers in the shortest possible time. Today's business environment is more competitive than ever before. As a consequence, e-business has spread to the mainstream market and gained serious consideration from almost every business sector.; In this thesis we study the e-business infrastructure along with its users as one system, techniques for availability and performance enhancement and their impacts on the end users. We study the Web user perceived Webserver availability by building and solving MRGP (Markov regenerative process) models. In the analysis we show the differences between the availability users see and the real system availability, and how to improve the former efficiently. Since software rejuvenation is often introduced to improve Webserver availability, we also address its impact on the user-perceived availability by modeling a Webserver with rejuvenation and user.; We discuss a non-parametric algorithm to estimate Webserver session timeout durations, which are often set arbitrarily by system administrators. Cost functions are formed to incorporate the user's convenience and security factors.; Software rejuvenation is an important method to address software aging problems, and system restart is a very common rejuvenation technique. For humongous enterprise servers hosting many different services, box-level software rejuvenation only is inadequate. We study an availability optimization problem of two-level rejuvenation policies by constructing a semi-Markov model.; Another cost-effective method to meet the increasing demand for capacity and high availability is clustering. By grouping many servers behind a request dispatcher or load balancer, the system availability can be made higher than that of a single server. In this environment, rejuvenation for individual servers is important to deliver even high cluster availability. We propose and analyze two rejuvenation strategies, called fixed rejuvenation and delayed rejuvenation, for clustering systems under varying workload. By solving the SRN (stochastic reward net) models, we compare the throughputs achieved by systems under these two rejuvenation policies. (Abstract shortened by UMI.)
展开▼
