A Multi Objectives Fault-tolerant Scheduling Algorithm Based on Consecutive Mapping

摘要

Aimed at computational nodes failures effect on applications. Purposed a multi objectives task scheduling algorithm. The algorithm can schedule predecessor task graph. Applied task copy and pipeline technique to deal with computational nodes failures. When computational nodes are failed on platform, tasks can be copied, replica of all predecessors are mapped onto distinct compute nodes, each replica of task communicates to a unique replica of its successor node, every replica of current task is mapped onto computational nodes involved in replica of predecessor node. The mapping strategy can reduce systematic communication. Task scheduling algorithm using the strategy is better than other existing algorithm in terms of the two system indices of latency and numbers of computational nodes.