Reliability/availability modeling and prediction for e-commerce and other Internet information systems

This paper describes the use of standard reliability modeling techniques - Markov Modeling and Reliability Block Diagrams - to analyze a web site and develop the answers to strategic questions on the configuration and operation of high availability computing systems. The analyses are performed using MEADEP, a powerful reliability analysis tool capable of hierarchical modeling and integrating Markov and block diagram techniques. In the example 3-tier architecture e-commerce site described in this paper, it is shown that (a) the most frequently failing subsystem is not necessarily the availability bottleneck, and (b) that restoration time is often a more important parameter than availability when attempting to maximize system throughput.