MRAFT: A New Fault Tolerance Algorithm (NOT PUBLISHED)

MRAFT (Modified RAFT) is an algorithm for achieving fault tolerance in multiprocessor systems. It is based on repeated executions and comparisons using duplex processors. Analysis given in this paper shows that MRAFT is equally efficient and reliable compared to its predecessor, RAFT (Recursive Algorithm for Fault Tolerance. ) Since MRAFT does not require storage of results from failed processors, a fault tolerant multiprocessor system could be configured using off-the-shelf processors and minimal additional hardware. It is shown that the system reliability of NRAFT environment is comparable to those of RAFT-based systems. For systems using VLSI processors with high reliability, MRAFT's performance is comparable to that of non-redundant operation and at the same time protection against intermittent faults is guaranteed by the on- demand time and space redundancy it provides.