Efficient Distributed Snapshot Algorithms with Less Message Overhead

Distributed snapshot algorithms to capture the consistent global states of a distributed system are presented. In addition to the state variable (red or white) of a process, the snapshot sequence number is also used to detect the messages crossing the snapshot line and hence, the messages which may cause the inconsistency of the snapshot line are easily detected. As a result, relatively low message overhead is required for the snapshot coordination. We assume non-FIFO communication channels and the normal operations of the processes including message sending and receiving are allowed during the snapshot coordination. The algorithm which includes only the processes which have communicated before the snapshot coordination is also presented.