PhD Thesis Defence


"Static and Dynamic View Selection in Distributed Data Warehouse Systems"

By

Mr. Panagiotis Kalnis


Abstract

Effective decision-making is vital in a global competitive environment
where business intelligence systems are becoming an essential part of
virtually every organization. The core of such systems are data
warehouses, which support On-Line Analytical Processing (OLAP) operations.
These are complicated statistical queries operating on large amounts of
historical data and should be efficient enough to allow interactive usage
of the system.

View materialization techniques are commonly used to accelerate OLAP.
Views may be materialized at any of the three layers of a decision support
system: (i) at the server side (i.e. the data warehouse), (ii) at a
mid-tier, between the server and the client and, (iii) at the client side.
Typically, static view selection is used at the server side, while dynamic
approaches are employed for the other two cases.

This thesis provides novel insights on view materialization techniques at
all three layers:

For the server-side layer, we discovered that the existing methods for
selecting a set of views to materialize have high execution cost that
renders them inapplicable to large problems. Motivated by this fact, we
propose the application of randomized local search algorithms, which
provide near-optimal solutions in limited time and are therefore useful
for real-life warehouses.

Since a data warehouse is typically accessed simultaneously by many
clients, we also investigated multiple-query optimisation methods, which
consider the available materialized views. Our methods exhibit
considerably better scalability than the previously known ones.

For the mid-tier, we propose a dynamic view materialization system, which
is based on cooperative OLAP Cache Servers (OCS). An OCS is the equivalent
of a proxy-server for web documents, but is designed to accommodate data
from warehouses and support OLAP operations. The OCS architecture is
beneficial for ad-hoc, geographically spanned users who access data
warehouses through the Internet. This approach requires a dedicated
infrastructure of OCS servers. We also investigate the alternative of
enhancing common web proxy-servers with OLAP capabilities by means of
active caching.

Finally, at the client side, we propose a similar system, but the data are
materialized locally at the clients instead of a mid-tier. By employing
Peer-to-Peer technology, we publish the stored data of each client, in
order to create a large virtual cache. Participation is ad-hoc, the system
is fully distributed and supports adaptive  reconfiguration. The
experimental results indicate that our system amplifies the benefits of
traditional client-side caching.


Date:			Friday, 18 January 2002

Time:			3:00p.m.-5:00p.m.

Venue:			Room 1505
			Lifts 25-26

Chairman:		Prof. Yifan Han (BICH)

Committee Members:	Dr. Dimitris Papadias (Supervisor)
			Prof. Frederick Lochovsky
			Dr. Qiang Yang
			Dr. Karl Lang (ISMT)
			Prof. Beng Chin Ooi (Comp. Sci., National Univ. of Singapore)


**** ALL are Welcome ****