http://www6.informatik.uni-erlangen.de:1200/Research/dist_sys.html (Einblicke ins Internet, 10/1995)
Horst Lührsen
Data Management in Distributed Systems
Description
The distributed Data Management System DDMS
was developed in 1989 by Stefan
Jablonski at the University of Erlangen Nuremberg. The System
is based on the Need-To-Know principle which means that data is not
necessarily made available as soon as possible at each node in a
distributed system but still fulfills the requirements of the applications
using these data. A data object needed by a certain application is allocated
locally at the applications node. Replicated objects are coupled by
propagation rules which allow synchronous as well as asynchronous
data propagation.
Current research focuses on different techniques of weak consistency
replication. The specification of propagation rules as described above is
one possibility of specifying concistency requirements. The main
disadvantage of this method is, that convergence of replica has to
be explicitly specified and is not enforced by any replication control
protocol. If consistency requirements are specified locally from an
applications point of view this disadvantage can be avoided. The idea
is to derive propagation rules as well as replication control protocols
from the local specifications. These protocols should preserve consistency
in a way that allows derivation of the actual value of any logical data item
at any time. This property is very useful for automatic recovery. However,
automatic recovery is not always possible no matter which specification
method is used. In these cases additional measurements have to be specified
in order to handle failures automatically.
More information
-
Anwendungsabhängige Spezifikation von
Konsistenzanforderungen bei replizierter Datenhaltung.
-
This report compares different techniques of application-oriented
consistency specification.
-
ASPECT - Specifying Consistency Requirements for Replicated Data from
an Applications Point of View.
-
This report desribes the ASPECT technique of specifying consistency
requirements for replicated data. ASPECT means Application-oriented
SPECificaTion of consistency requirements. The name ASPECT was also
chosen to indicate the underlying need-to-know principle: The Data
Management System provides an application only with those aspects
of data needed by the specific application.
Contact
Richard Lenz
lenz@informatik.uni-erlangen.de