| This dissertation investigates the reliability modeling and evaluation in computer networks and distributed systems. High performance computer networks and distributed systems are essential for many real-world applications. Now that those systems have become affordable and widespread, reliability is becoming one of their most important attributes.;In conventional network reliability, a n-node computer network is modeled by an undirected probability graph whose vertices represent the nodes (assumed not to fail, i.e. perfect), and whose edges represent the links (assumed to fail, i.e. imperfect). Therefore the network reliability measure is based upon the reliability measures of the individual links. Three known measures of network reliability are the all-terminal, 2-terminal, and k-terminal reliabilities. All-terminal reliability requires that every node to be able to communicate with every other node in the network. Two-terminal reliability requires that a distinguished pair of nodes, the source and the destination, be able to communicate with one another. K-terminal reliability requires that a specified set of k target nodes be able to communicate with one another.;Hypercube networks have received much attention in recent years as vehicles for concurrent processing. Assume that N = 2k for some k ≥ 0, and let N processing nodes P0, P1, ..., P N-1 be available. A k-dimensional hypercube network is obtained by connecting each processing node to exactly k neighbors. The conventional hypercube network reliability has been reported, but the reliability of hypercube networks with imperfect and/or spare nodes has not received much attention. This dissertation first presents an analytic model for the reliability evaluation of those hypercube networks.;Hardware reliability and software reliability in distributed systems are then investigated. For hardware reliability, this dissertation focuses on the distributed multimedia systems with the reliability modeling and evaluation of entertainment switching units, entertainment servers, and video/audio broadcasts. The reliability model is validated by comparing analytic results with simulation results.;For software reliability, reliability of distributed recovery blocks in real-time distributed systems and reliability of remote procedure calls in distributed systems are extensively studied. Finally, this dissertation also includes conclusions regarding the obtained results and suggestions for future research. |
