Self-Organizing Intelligent Data
PhD
Research Programme
Ali Adams
Bournemouth University, UK. 1996-2000
Overview
Information
resources are currently managed using database technology. Users access
information on database management servers using client applications, and
server administrators control client access to the information using
authentication, views, etc. Market globalisation trends are leading
organisations to collaborate and share information resources for mutual
benefits, but yet still, retain ownership of their individual part of the
collaborating information resources. Naturally, a huge growth in the number of
users of the system follows, and an exponential increase in concurrent data
traffic between clients and servers is inevitable. To overcome this problem,
either high bandwidth communication links need to be used, or data reallocation
needs to be carried out dynamically to cope with continuous changes in
information access patterns.
This work
proposes a new data management model for distributed information systems that
assumes no global knowledge of either data schema or network topology, and as
such, it allows for the creation of flexible and scalable information systems
that can inherently cope with the above requirements.
The main aim is
to identify the intelligence requirements a data object may need, (how to acquire,
use, and adapt knowledge), in order to be able to exhibit
intelligent self-organization behaviour (partition, migrate, replicate, and/or
combine) in response to dynamic changes in demand patterns and environment
resource usage. In general, data will acquire and use self-organization
knowledge using fuzzy logic, and adapt that knowledge (learn by experience)
using genetic algorithms.
Related data
objects are packaged into autonomous data sources capable of exhibiting
self-organisation behaviour, the goal of which is to improve overall system's
performance. By monitoring their access patterns and system resource
distribution, data sources can decide to partition, migrate or replicate near
their users, and/or combine with other data sources.
Using simulation,
it is shown that significant performance improvements over conventional systems
with no dynamic relocation can be achieved when using such a data
self-organisation model. Experimental results showed that different types of
self-organisation actions contribute to the improvements to varying degrees.
Migration, while contributing the most under typical operational conditions,
replication can contribute significantly more for systems with low levels of
update queries. On the other hand, partitioning and combining, although does
contribute to the performance improvements, their cost out-weigh their
contribution and hence they are kept to minimum.
More
importantly, collective emergent self-organisation behaviour has been observed
which shows an increase in the locality of reference for data access. This
directly translates into significant reductions in both data access time and
network traffic volume.
Please download
the complete thesis document and feel free to request
more info or send feedback to me.
ThanQ.
Ali Adams
GUI Design Consultant
[email protected]