Bi-level reconfigurations of fault tolerant arrays

Two types of algorithms are considered, namely, local algorithms and global algorithms. In a local algorithm, no processors need to know the status of all other processors in the system. The recovery process is distributed among the processors with each processor using extremely local knowledge. With these properties, the reconfiguration algorithm may achieve fast recovery and real time response but many sacrifice the optimal use of redundancy. In contrast, the goal of a global algorithm is to optimize the use of redundancy with respect to some fault tolerance criteria. This, however, requires global knowledge about other processors in the system and often necessitates extensive changes in the configuration of the system. For unmaintained, long-life systems, local fault tolerance algorithms have the advantages of fast recovery, while global fault tolerance algorithms provide better reliability and longer life expectancy. Fortunately, under certain conditions, it is possible to combine the advantages of the two types of algorithms. These conditions are described.<
>