Development and Performance Analysis of a Fault Tolerant Algorithm for Cluster of Workstations

A Cluster of Workstations (COW) is network based multi-computer system, which is the most prominent distributed memory system aimed to replace supercomputers. A cluster of workstations can be viewed as a single machine in which one job is divided into n subtasks and delegated to n workstations in the COW architecture. To get the job completed, all subtasks assigned to component workstations must be completed. Therefore, for satisfactory job completion, all workstations must be functional. However, a faulty node can suspend the over all job completion task until. Therefore, a job can not be completed until a faulty node is recovered from fault. This paper presents a fault tolerant architecture for COW, which will allow a normally working workstation to perform the tasks of the faulty workstation in addition to its original assignments. The Markov models are basic tools applied for availability modeling. This paper presents a Markov Availability model for estimating the availability of component workstations as a function of workstation failure rates.