2011 IEEE 13th International Symposium on High-Assurance Systems Engineering (2011)
Boca Raton, Florida USA
Nov. 10, 2011 to Nov. 12, 2011
DOI Bookmark: http://doi.ieeecomputersociety.org/10.1109/HASE.2011.14
Many fault-tolerant systems organize the replicas of an application process as a process group. The Leader-Determined Membership Protocol determines a new membership for the process group, when a member becomes faulty, a member leaves the group, or a new member joins the group. If the primary becomes faulty, the protocol selects a new primary deterministically, based on the precedences and the ranks of the backups. The new primary determines which processes are members of the new membership, and communicates that information to the backups. The protocol maintains a consistent view of the membership, so that the members see the same primary, the same set of members, and the same primary view number. It also ensures consistency of the states of the members, and consistency with other processes with which the members communicate. By avoiding the use of a multiple-round majority-based consensus algorithm, the Leader-Determined Membership Protocol achieves better performance than other membership protocols, and can operate with the primary and only a single backup.
distributed systems, fault tolerance, membership, process group, replication, virtual synchrony
W. Zhao, L. Moser and P. Melliar-Smith, "Leader-Determined Membership Protocol," 2011 IEEE 13th International Symposium on High-Assurance Systems Engineering(HASE), Boca Raton, Florida USA, 2011, pp. 124-129.