Fault-tolerant control for Scalable Distributed Data Structures

Abstract

calable Distributed Data Structures (SDDS) can be applied for multicomputers. Multicomputers were developed as a response to market demand for scalable and dependable but not expensive systems. SDDS consists of two components dynamically spread across a multicomputer: records belonging to a file and a mechanism controlling record placement in the file. Methods of making records of the file more or less fault-tolerant have already been given. Methods of making the mechanism controlling record placement in the file fault-tolerant have not been studied, yet, although it seems that this is more important for the system dependability than record fault tolerance. Faults in control may lead an application to crash, while record data faults may cause invalid computations at most. In the paper a fault-tolerant control for SDDS is given. It is based on an application of Job Comparison Technique along with TMR. Time overhead due to redundancy introduced is estimated, too.