一种低开销非阻塞的协同式检查点算法

协同式检查点设置及卷回恢复技术是一种简单有效的容错手段，被广泛地运用于并行/分布式系统中。为进一步降低协同式检查点算法的开销，该文给出了一个基于可重建检查点的非阻塞协同式检查点算法。并行程序出错导致卷回恢复发生的概率远小于检查点设置概率，该算法利用这一特性，将检查点设置的部分开销转至卷回恢复阶段，降低了容错的开销，提高了系统的可扩展性。

Coordinated Checkpoint Algorithm of Low-overhead and Non-blocking

As an effective method of fault-tolerance, technologies of coordinated checkpoint and rollback recovery are widely used on the parallel or distributed computer systems. In order to reduce the overhead of checkpoint time, this paper proposes a low and non-blocking coordinated checkpoint algorithm based on reconstructed checkpoint. Checkpoint happens much more often than rollback, fractional consumption of checkpoint setting is turned to rollback recovery stage. The algorithm lowers fault-tolerance consumption, and improves system’s scalability.