A methodology for comparing fault tolerant computers

A general methodology for comparing fault tolerant computers using a hierarchical taxonomy of pertinent issues is described. The uppermost level taxonomy consists of reliability, specialized fault detection, performance, flexibility, and communication issues. Each class is further decomposed into more specific topics. The methodology consists of studying each issue in turn to determine which candidate architecture(s) best support that issue. The results are then combined using a user defined function to rank the architectures. A design space technique is described to aid in drawing the comparisons. The technique is illustrated by considering the fault tolerant capabilities of the JPL hypercube and the Encore Multimax.<
>