Bounding rollback-recovery of large distributed computation in WAN environment

In the existing optimistic message logging protocols, the dependency must be tracked in whole system, and all processes are involved in rollback recovery in the event of failure. For large distributed computation in WAN environment with the low available bandwidth and high transmission latency, its fault-free overhead and recovery overhead are outstanding, recovery efficiency decreasing with the scale of system. This paper introduces a three-layer model of large distributed system in WAN environment, and presents a protocol of message dependency tracking based on proxy. Utilizing private proxy to log messages and dependencies, the protocol limits rollback-recovery to a scope called block rather than the entire system, achieving relative low fault-free overhead and fast output commit, as well as improved recovery efficiency and low recovery overhead.