The t(n-1)-diagnosability and its applications to fault tolerance

Abstract

A system composed of n units is said to be t/(n-1)-diagnosable if, given any complete collection of test results, the set of faulty units can be isolated to within a set of at most n-1 units provided that the number of faulty units does not exceed t. Based on some recently discovered properties of t/(n-1)-diagnosability, the author examines three canonical classes of systems-chains, loop and H/sub 2r,n/ systems-and presents optimal t/(n-1) diagnosable configurations for these classes. Incorporating these results into the scheme of D.M. Blough and A. Pelc (see 20th Inst. Symp. on Fault-Toler. Computing, pp.316-323 (1990)), the author gives an improved diagnosis and repair algorithm for constant-degree multiprocessor systems. A software fault tolerance scheme that utilizes t(n-1)-diagnosis technique is also proposed.<>