Survey on system level fault diagnosis

With the advent of inexpensive microprocessor elements the design of large multiprocessor computing networks becomes feasible. However, there are many interesting and important problems which must be solved before such system can be utilized in a meaningful way. These problems involve many aspects of computation,system structure and reliability.Much work is in progress to derive significant trade offs among the following requirements: from the point of view of computation,the development of parallel algorithms which allow the full exploitation of the intrinsic level of parallelism which is present in large computing networks; the devolopment of new architectures with prominent attention on the of the intercommunicationst network, which has become one of the main parameters in the overall cost of a system.fi- nally the development of efficient tools which allow an early detection of faults, an'automatic diagnosis of fa ulty units with consequent automatic reconfiguration of the system and recovery of the programs to safe operation. Since a system is configured as several computationally and logically interconnected units, tentative solutions to the previous problem must be derived from a system level approach rather than from a logic circuit level approach. In the field of automatic diagnosis, formal model has been developed and subsequently a generlized which allows the identification of desirable cha- racteristics in the design of a system for a system le- vel self-diagnosis which, is the prerequisite of an auto matic and user transparent reconfiguration and recovery. In this paper, these results will be surveyed and expected trends will be discussed.