Athens University Economics and Business (AUEB)
Techniques and systems for online and distributed database processing
Abstract
dc:descriptionModern information processing is moving into a realm where we often need to process large amounts of data on daily basis, as a result of increasing monitoring of environmental parameters, market transactions, web activity, and other sources of valuable information. The availability of this data establishes new business and application scenarios with increasing performance requirements. Current database engines are expected to analyze the data while providing a high query throughput. In order to accommodate user needs, it is necessary to reconsider the role of several database components in a new light. In particular, advances of the state of the art need to be introduced in the query processing, transaction processing and index management areas. Advances in the query processing module of the database engine are triggered by the need to accommodate new scenarios. While traditional query processing is designed with the assumption that data is always available, modern database systems should be ready to process various types of incoming, streamed data, often arriving at uncontrollable rates or experiencing variable delays. New query processing algorithms, and in particular join algorithms, need to be designed in order to retrieve valuable information from this type of input as soon as incoming data is available. This thesis addresses two common operations: the join of two (or more) data inputs streamed in by autonomous sources and the join of a data stream with a disk-resident relation. The transaction processing system is nowadays faced with new challenges as a result of data replication. Data replication introduces the potential for higher availability and better performance compared to the traditional centralized management, by sharing the database load between several engines. However, maintaining consistency amongst the replicas in the presence of updates is far from trivial. Transaction processing needs to be governed by novel concurrency control mechanisms suitable for data replication. This thesis shows how to provide transactional properties, atomicity, consistency, isolation and durability, in a replicated database in an efficient and scalable manner. Selecting indices that are appropriate for the workload is an important task for the database, since indices directly impact the performance of query processing. It is also an interesting optimization problem, as indices may help some parts of the workload, while requiring maintenance overhead when the data is updated. As database applications become increasingly complex, index tuning becomes more and more challenging for database administrators. This thesis introduces an online technique that captures the relevant changes in the workload and manages the indexes in parallel with query processing.
Degree
thesis:*- Grantor dc:publisher
- Athens University Economics and Business (AUEB)
- Year dc:date
- 2010
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Bornea, Mihaela-Ancuta
Subjects
dc:subject × 12Rights
- Language dc:language
- eng
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/22720