Fault-tolerant task management and load re-distribution on massively parallel hypercube systems

The authors present a scheme for managing real-time task allocation and load redistribution with fault-tolerance for hypercube systems. A set of processors, called fault-control processors (FCPs), can be used for keeping the duplicate copies of tasks and real locating tasks if the original processors of those tasks fail. Two-level task redundancy is used by grouping the FCPs as primary and secondary for each processor. The proposed scheme provides a high degree of fault-tolerance since each FCP itself is monitored by other FCPs. Assuming a failure-repair system environment, the performance of the proposed strategy has been evaluated and compared with a fault-free environment for 256-node and 512-node hypercubes, through simulation experiments. The authors also introduce a measure of goodness, success probability, which represents the probability of reallocated tasks meeting their deadlines despite the failures of processors. It is shown that, using the proposed scheme, a large percentage of the rescheduled tasks can still meet their deadlines. The probability of a task being lost altogether, due to multiple failures, has been shown to be extremely low