A non-FIFO checkpointing protocol for distributed systems

The checkpointing and rollback recovery are well-known techniques that allow processes to make progress despite of failures. The design of an effective checkpointing protocol in the coordinated approach requires every process to checkpoint its state periodically. A major difficulty is raised due to the fact that processes cannot checkpoint their states at exactly the same instant in a distributed system where there is no common clock or clocks cannot be kept perfectly synchronized. The authors present a checkpointing protocol that does not require the communication subsystem to deliver messages in a first-in first-out fashion