Summary: | 博士 === 國立清華大學 === 資訊科學學系 === 83 === Quorum-based algorithms are an important class of algorithms to
achieve distributed mutual exclusion. They are resilient to
network partitioning caused by site and/or network link
failures and usually evoke low communication cost. The basic
idea of them is simple-a site should collect permissions
(votes) from all sites of a quorum to enter the critical
section. If we can assure that any pair of quorums have a non-
empty intersection and each site gives its permission to only
one site at a time, mutual exclusion is then guaranteed. The
collection of quorums used by a quorum-based algorithm is
called a quorum structure. According to different mutual
exclusion scenarios, several types of quorum structures have
been proposed: coterie, wr-coterie and k-coteries, which are
related to distributed mutual exclusion, replicated data
consistency and distributed k- mutual exclusion, respectively.
In this dissertation, we propose novel methods for constructing
coteries, wr-coteries and k-coteries that are nondominated and/
or of constant expected quorum size. The proposed methods can
easily be extended to solve the problems of mutual exclusion,
replicated data consistency or k-mutual exclusion in a
distributed system. Nondominated quorum structures are
favorable because they are candidates to achieve the optimal
availability, the probability that a quorum can be form in an
error-prone environment. On the other hand, quorum structures
of constant expected quorum size are preferable because when
the proposed methods are applied to solve the problems
mentioned, the message cost is directly proportional to the
quorum size.
|