Antecedence Graph Approach to Checkpointing for Fault Tolerance in Multi Agent System

Checkpointing has been widely used for providing fault tolerance in multi-agent systems. But the traditional message passing based checkpointing and rollback algorithms may suffer from problems of excess bandwidth consumption and large overheads. In order to maintain consistency of multi agent system, the checkpointing is forced on all participating agents that may result in blocking of agents’ operations to carry out checkpointing. These overheads could be considerably reduced if the checkpointing would be forced only on selective agents instead of all agents. This paper presents a low latency, non-blocking checkpointing scheme which marks out dependent agents using Antecedence graphs and then checkpoints are forced on only these agents. To recover from failures, the antecedence graphs and message logs are regenerated and normal operations continued. The proposed scheme reports less overheads and reduced recovery times as compared to existing schemes.