Design of fault tolerant multiprocessor systems

Abstract

The purpose of this thesis is towards designing a Fault-Tolerant Multi Processor newlineSystem useful for space or military applications. INMOS T800 has been conceived on newlinethe basis of building block for multiprocessing, because of its link based architecture. newlineTransputer is inherently more Fault-Tolerant than any other standard bus based newlinearchitecture. newlineThe multiprocessor network design is developed around and#8213;Triple Modular newlineRedundancyand#8214; philosophy with autonomous decentralized loop configuration for newlinenetworking. The powerful fault-tolerant nature of the network is gradually unveiled as newlineone goes through the thesis. The concurrent programming language OCCAM - 2 is used newlinefor real time programming. The programs, examples presented in this thesis were newlineimplemented using OCCAM programming language in Transputer Development System (TDS). The sources of different kinds of faults have been discussed and also various newlinehardware and software schemes towards achieving fault-tolerance and hence overall newlinesystem reliability are outlined in the thesis. Fault tolerance is generally achieved through newlinethe use of redundancy of hardware and/or software (spatial redundancy), of time newline(temporal redundancy), of information (value redundancy), or a combination thereof. newlineSpatial redundancy means that individual elements of hardware and/or software that can newlinebe faulty are replicated. There exist three types of spatial redundancy: static, dynamic newlineand hybrid. newlineStatic redundancy is a technique that works by and#8213;maskingand#8214; faults. This is newlineperformed by executing a distinct task on each of the replicated elements. Dynamic newlineredundancy involves two phases. In the first phase faulty elements of the system are newlineidentified in some manner. In the second phase the system recovers from the faulty units newlineby switching over their operation to redundant, backup elements that are hopefully available and fault-free. The success of this method relies on choosing a suitable number newlineof spares, an effective technique for detecting faults, and a viable switching operation. newline