|
Abstract
Existing research on replication is generally based on
synchronous replication (all copies are kept consistent at all times) and
update everywhere (any replica can be updated) approaches. There is a
strong belief among database designers, however, that synchronous, update
everywhere replication is simply not feasible in a database environment.
As a result, most current database replication solutions are asynchronous
(copies are not kept consistent at all times) and based on a primary-copy
approach (only one master copy can be updated, all other replicas are
read-only). One of the reasons for this is that many of the existing database
replication protocols ignore the complexities and overhead of the
communications involved. Group communication primitives have been
suggested as a way to solve this particular problem and to provide the support
required to implement synchronous, update everywhere replication.
This approach has considerable potential, its main drawback being that
there is still a big gap, both conceptual and practical, between
group communication protocols and databases. Although
some work has been done to bridge this gap, a significant number of questions
remain open. One of them is how to support partial replication.
|