分布式仿真系统非阻塞式协同检查点协议

联邦存储是实现分布式仿真系统容错的基础,IEEE1516-2000中提供了一种阻塞式的系统状态存储协议,实际使用时会给系统带来很大开销。根据对系统状态存储过程中仿真组件状态变化影响因素的分析,提出了一种非阻塞式的系统状态存储协议,该协议完整地解决了由于在系统状态存储期间允许时间推进导致所保存的RTI状态和对应成员状态时间上的不一致、由于成员间消息传递导致不同成员状态的不一致及in-transit消息问题。先完成状态保存的成员不需阻塞,该协议在一定程度上降低了联邦存储的开销。证明了该协议的正确性。

Non-blocking Coordinated Checkpointing Protocol for Distributed Simulation System

Federation save is the foundation of fault tolerance of distributed simulation system.HLA standard,IEEE 1516-2000,provides a blocking coordinated checkpointing protocol for saving federation state,which brings huge overhead to simulation system.Based on the analysis of operation mechanism of RTI,a non-blocking coordinated checkpointing protocol was put forward.The protocol solved time inconsistency between the saved RTI state and corresponding federate state due to time advance request and state inconsistency among different federates due to message transmission,and in-transit messages were also treated.By this protocol,the federates that finished state save ahead of others don't need to wait for the ending of the federation save,and the overall system state store overhead decreased.The correctness of the protocol was proved.