一种基于索引的准同步检查点协议

An Index-Based Quasi-Synchronous Checkpointing Protocol

在基于索引的分布式检查点算法中,尽量减少全局一致性检查点和强制检查点的数目对提高计算效率具有重要意义.该文在已有的基于索引的检查点算法的基础上,提出了一种新的检查点协议,既减少检查点的数目,又使各个进程的检查点之间实时同步,以免程序出错后回卷执行的开销太大,丢失过多有效计算.模拟实验表明,按该文所提协议,平均每条消息导致的强制检查点数比传统方法平均减少23·2%.

To provide rollback-recovery for fault-tolerance in distributed systems, it is significant to reduce the number of checkpoints under the existence of consistent global checkpoints in indexbased distributed checkpointing algorithms. A new checkpointing protocol is presented in this paper on the basis of index-based checkpointing protocols. It not only reduces the number of forcedcheckpoints but also keeps synchronous in time to avoid too much amount of overhead of roll-back recovery due to useful computation losing in case of failure. Simulation results show that the proposal algorithm in this paper can reduce the number of induced forced-checkpoints per message 23.2 % on an average comparing to the traditional strategies.