Method and apparatus for a checkpoint based communication processing

Abstract

A checkpoint based communication processing system and method in which an operation is executed and during execution checkpoints are periodically taken in order to recover from a failure which may be detected in the operation between any two checkpoints. Communication processing is controlled in accordance with a send-to-request instruction from an application. A plurality of communication packets for the communication processing are generated and a sequence numbered for a checkpoint is incremented each time when a checkpoint has periodically been taken during the operation. Affixed to each communication packet is attribute information in the form of the checkpoint sequence number existing at the time of generation of the communication packet, with plural communication packets generated between incrementing of the sequence number having affixed thereto the same sequence number. Whether or not the communication is promptly transferred is determined based on a comparison of the fixed attribute information and the latest sequence number. In another embodiment, the attribute information indicates whether or not the packet is the last of the generated of the plurality of transfer packets, and transfer packets fixed with attribute information which does not indicate the last position is transferred immediately, but the transfer packet which has affixed attribute information indicating the last position is delayed in transmission until a next checkpoint has been taken.