Reliable distributed sorting through the application-oriented fault tolerance paradigm

A fault-tolerant parallel sorting algorithm developed using the application-oriented fault tolerance paradigm is presented. The algorithm is tolerant of one processor/link failure in an n-cube. The addition of reliability to the sorting algorithm results in a performance penalty. Asymptotically, the fault-tolerant algorithm is less costly than host sorting. Experimentally it is shown that fault-tolerant sorting quickly becomes more efficient that host sorting when the bitonic sort/merge is considered. The main contribution is the demonstration that the application-oriented fault tolerance paradigm is applicable to problems of a noniterative-convergent nature.<
>