SFT:一个具有较短冻结时间的一致检查点算法

介绍了一个基于消息记录的一致检查点算法－ＳＦＴ算法，ＳＦＴ算法能够实现分布式系统的容错，该算法具有无多米诺效应，冻结时间短，开销小和重启动算法简单的优点，ＳＦＴ的ＩＰＣ机制基于ＰＶＭ，能够保证消息的有序到达，并且其消息的发送和接收操作都是原子操作，另外，ＩＰＣ机制中进程的ｉｄ值编码与所在机器无关，这样一个过程即使从故障机器迁移到其它机器上运行仍可与其它进程继续通信，为提高检查点操作的并行性，ＳＦＴ

SFT: A CONSISTENT CHECKPOINTING ALGORITHM WITH SHORT FREEZING TIME

SFT algorithm, a consistent checkpointing algorithm based on message logging, is presented in this paper. SFT is able to implement fault tolerance in distributed systems. The algorithm has such distinguishing features as, domino free, short freezing time, low overhead, and roll back simply. In SFT, message send and receive operations are atomic operation. And just like PVM, messages are received in the same order as they are sent. Moreover, unlike PVM, in SFT a process's id has nothing to do with its current running machine. So even if a process' checkpoint file continue running on another machine after recovery from an error, it also can send and receive message in the same way. By using main memory algorithm and no global synchronization needed, the algorithm has a high concurrency. To make SFT's freezing time shorter, a special control message, Munblock message, is used to ensure that a process could respond the checkpoint event quickly at any given time. In SFT, even when a process is waiting for receiving an incoming message, it also can respond the checkpoint signal and do its local checkpoint at once. Furthermore, the control messages' number of SFT algorithm is only O(n) .