A reliability study on switching fabric based system architecture

Processor and networking technologies have outrun system technologies, such that the system itself is the bottleneck. Many vendors and organisations are trying to address this problem by replacing the popular bus based architecture with a switching fabric. The paper studies the new switching technology and compares it with the old bus based architecture. Both architectures are to be used in a highly-available parallel processing environment. It is assumed that both systems have incorporated the same fault tolerant strategies. A hierarchical approach is used to break down the systems into subsets, such that both software and hardware components are considered. Each component is represented by a hazard function or a failure intensity. The fault tolerant strategies are used to estimate the repair time probability distribution function. The availability of the system is calculated. It is found that the switching fabric has a slightly higher availability.