A crash recovery technique in distributed computing systems

In this paper we propose a new mechanism for implementing checkpoint/rollback-recovery in a distributed computing system. A logical-ring structure is introduced for the maintenance of recovery-related information. Message processing order of a process is maintained by all other processes on its associated ring. It requires no time-consuming operations of writing order information into stable storage. As a result, fail-free overhead is small. When failures occur, only failed processes have to roll back to their latest checkpoints. Surviving processes continue execution without being blocked. Output commit is fast as it needs no synchronization before a message is sent to the outside world