| Summary: | 博士 === 國立臺灣大學 === 電機工程學研究所 === 94 === Data cube has become an important component in most data warehouse systems and in decision support systems. Modern data warehouses have a huge amount of data, and OLAP queries submitted by users are becoming more complex. In this dissertation, we first devise the mechanism for striking the balance between response time and the cost of storage space.
Then, we propose a framework of approximating query processing for data
cubes. Finally, we extend the framework to deal with cube streams.
An ideal OLAP system is expected to provide acceptable response time,
controllable updating cost, and least storage space. To guarantee a
satisfactory query response time, the pre-computed techniques, also known as
view materialization, are developed. Materialized Views (MVs) have been
found to be very effective in speeding up query as well as update
processing, and the methods are being widely supported by commercial
database systems.
In addition, users usually pose very complex queries to the OLAP system in
recent data warehousing systems, which requires complex operations over
gigabytes of data and takes a very long time to produce exact answers.
Consequently, the issue of approximating OLAP queries becomes critical.
Answering range queries is one of the primary tasks of OLAP applications.
However, datasets tend to be very large in real data warehousing systems.
Thus, answering aggregate queries can be computationally expensive. To
address this issue, providing approximate answers to online queries is a
viable solution. Also, error bound estimation of the answers to queries is
an important functionality for users. That is, both an efficient approximate
query processing algorithm and estimation of error bounds are required for
DSS applications.
For multidimensional data streams, or cube streams, the volume of data is
usually too huge to be stored in permanent devices or be scanned thoroughly
more than once. Such applications have to process cube streams with limited
resources and keep the approximated information in a synopsis memory for
further analysis. In addition, the resources for both the processing time
and the memory are much more constrained than in off-line cube construction
so that cube streams must be processed efficiently with a small working buffer. As a result, an efficient algorithm that can compress cube streams within a small working buffer in one data scan is required to address such a
problem.
In this dissertation, we proposed the solutions for those issues of modern
OLAP systems. First, we devise an efficient mechanism to solve the problem
of how to arrange materialization tasks, and propose the algorithm MAVIS,
striking the balance between response time and the cost of storage space.
Second, the framework DAWN, focusing on answering range-sum queries with
error estimation from compressed OLAP cubes, is proposed. Third, we propose
the DAWA algorithm, an integrated algorithm of Dct for Data and discrete
WAvelet transform, for approximating the cube streams with very restricted
resources.
With the the framework DAWN, algorithms MAVIS and DAWA, data warehousing systems are able to answer OLAP queries efficiently with limited storage space. In addition, OLAP systems are able to answer range-sum queries from
compressed data cubes instead of dealing with huge volume of data cells.
Moreover, the multi-dimensional data streams, or cube streams, could be
processed and stored very efficiently.
|
|---|
