一种基于高性能集群计算系统的检查点策略

为了提高高性能集群计算系统的容错能力,检查点设置成为一种广泛采用的手段.目前检查点设置多采用的协调式设置协议,该协议在集群规模扩展情况下,同步操作造成巨大的系统时间开销,并阻塞正常计算的执行.针对该问题,使用非协调式检查点设置协议消除同步操作,采用消息日志记录方式保证系统状态一致性,并利用线程后台执行方式达到透明性设置.最后,通过典型的系统实验,验证了该方法的有效性,并进行同协调式协议设置的时间开销对比.

Checkpointing Strategy for High Performance Computing Cluster System

Checkpointing is an effective way of improving the reliability of cluster system. It's a research hot spot how to lower the system spending of checkpoint. At present, the study of checkpointing points that the rollback recover of un-coordinated checkpointing is difficult to realize,and has less appilication in cluster system. This paper will compare existing schema use non-blocking checkpoint and reduce synchronization operation, so that system spending will be cut down, and the efficiency will be improved.