paper

A History of the Virtual Synchrony Replication Model

  • Authors:
Read the paper

📜 Abstract

In summary: virtual synchrony is a distributed execution model that guarantees a very strong notion of consistency. Applications can create and join groups (potentially, large numbers of them), associate information with groups (the state transferred during a join), send multicasts to groups (without knowing the current membership), and will see the same events in equivalent orders, permitting group members to update the group state in a consistent, fault-tolerant manner. Moreover, although we’ve described the virtual synchrony model in pictures, it can also be expressed as a set of temporal logic equations. For our purposes in this chapter, we won’t need that sort of formalism, but readers can find temporal logic specifications of the model in [71][23].

✨ Summary

Influence

  • The chapter is part of Replication: Theory and Practice, a 2010 Springer volume that surveys approximately three decades of replication research. Its subject matter was subsequently used as background material for teaching dynamic reconfiguration and view-synchrony concepts in distributed-systems courses. (link.springer.com)
  • Later research cites the chapter when relating virtual-synchrony-style reconfiguration to consensus protocols. For example, work on Stoppable Paxos explicitly references Birman’s chapter when discussing reconfiguration in which one configuration stops at a globally known state before another configuration begins. (theory.stanford.edu)
  • The chapter’s account of dynamic process groups also appears in subsequent fault-tolerant high-performance-computing research. A 2018 study of resilient MPI applications cites the chapter and identifies its Dynamic Group of Recommended Processes as being inspired by group systems such as Isis, particularly their self-managed membership model. (inf.ufpr.br)
  • The paper records the broader technical and industrial adoption of virtual synchrony through systems and standards including Totem, Transis, Phoenix, Spread, CORBA fault tolerance, and data-center management systems associated with IBM and Microsoft. These uses are documented in the paper itself and provide concrete evidence of the model’s role in fault-tolerant communication and service replication. (cs.cornell.edu)