Elsevier

Information Systems

Volume 18, Issue 5, July 1993, Pages 343-348
Information Systems

Designing computer networks to avoid partitioning

https://doi.org/10.1016/0306-4379(93)90033-WGet rights and content

Abstract

It is shown that a network can be constructed on a given set of host computers such that the possibility of a network partition resulting from network node and link failure can be ruled out with an arbitrarily high degree of confidence. More precisely, a class of networks is exhibited on any given number of host nodes so that the probability of a network partition decreases exponentially with only a linear increase in connectivity cost. It has long been a folk theorem in network theory that as one increases the budget for the number of links of a network, the reliability of the network can be increased by a judicious choice of network topology. This paper makes this intuitive statement precise and analyzes one class of networks to illustrate it: bipartite networks where host nodes are connected to each of a set of hub nodes. The result has significant implications for availability of distributed databases and feasibility of the three-phase commit protocol which guarantees crash recovery in distributed transactions.

References (9)

  • D Skeen

    Nonblocking commit protocols

  • D Skeen et al.

    A formal model of crash recovery in a distributed system

    IEEE Trans. Software Engng.

    (1983)
  • S Ceri et al.

    Distributed Databases: Principles and Systems

    (1984)
  • M Özsu et al.

    Principles of Distributed Database Systems

    (1991)
There are more references available in the full text version of this article.

Cited by (0)

View full text