INPUT SEQUENCE REORDERING METHOD FOR SOFTWARE FAILURE RECOVERY

A software failure bypass system based on checkpointing, input logging, rollback and input sequence reordering is disclosed. The software fault bypass system minimizes the number of involved processes as well as the total rollback distance. The software fault bypass system includes a progressive retry recovery algorithm which gradually increases the scope of the recovery when a previous retry step fails to bypass the detected fault and a reorder recovery algorithm which will attempt to bypass the detected fault by reordering and reprocessing the inputs that have been received by the faulty application process. The fault tolerant system includes a mechanism which utilizes information about the particular application process or the cause of the detected fault to determine if the recovery of a particular application process should proceed directly to the reorder recovery algorithm. In one embodiment, if the faulty process exhibits nondeterministic behavior, the overall progressive retry recovery algorithm should be utilized to recover the faulty process. If the faulty process does not exhibit nondeterministic behavior, however, the reorder recovery algorithm should be utilized to recover the faulty process.