Aristotle University Of Thessaloniki (AUTH)
Βελτιστοποίηση ερωτημάτων σε κατανεμημένα και δυναμικά περιβάλλοντα
Abstract
dc:descriptionAccording to a 2012 IBM annual report, the volume of the data that is produced every day exceeds the 2.5 quintillion bytes. The sources of this data vary from social networks and data sharing sites to sensor networks and mobile phones. The latter phenomenon has given rise to the development of novel infrastructure for storing and processing data, such as clusters, grids and clouds. The development of novel query/worfklow optimization and processing techniques is more than necessary in order to take advantage of the maximum of the potentials of this infrastructure as state-of-the-art algorithms tend to overlook many of the peculiarities of modern infrastructure. Thus, the produced plans are less efficient or even inconsistent with the user requirements. This thesis aims to develop algorithms and techniques for overcoming the limitations of state-of-the-art work. Its main contributions are summarized to the following: (i) A novel provably optimal algorithm has been developed for building query/workflow plans when the SQL operators/user tasks are executed through remote service calls. The goal is to minimize the response time of a query/workflow. The assumptions that were considered and differentiate the proposed algorithm from the state-of-the-art ones are, first, the adoption of pipelined parallelism and, second, the possibility of direct communication between remote services through heterogeneous network links. The motivation for studying this problem was the increasing adoption of services by modern infrastructure. (ii) Two generalizations of the previous algorithm have been also developed that account for multi-way join services. The cases where the proposed algorithms build the minimum response time plans have been also studied. (iii) A new adaptive technique has been introduced for query processing in dynamic environments. Experiments performed both with synthetic and real-world data show that the proposed technique builds more efficient plans incurring less runtime overhead. This work has been motiv! ated by the highly dynamic and unpredictable nature of modern infrastructure and data. (iv) Finally, motivated by the fact that the resources of modern infrastructure are leased, several algorithms have been proposed for query/workflow scheduling over multiple clouds. The latter algorithms aim to schedule queries/workflows considering both an SQL operator's/task's response time and the resource leasing cost. The difference between the proposed algorithms and the state-of-the-art ones is the assumption that clouds express their response time-resource monetary cost characteristics through continuous functions.
Degree
thesis:*- Grantor dc:publisher
- Aristotle University Of Thessaloniki (AUTH)
- Year dc:date
- 2013
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Tsamoura, Efthymia
- Τσαμούρα, Ευθυμία
Subjects
dc:subject × 12- Κατανεμημένη βελτιστοποίηση ερωτημάτων
- Προσαρμόσιμη βελτιστοποίηση ερωτημάτων
- Πολυκριτηριακή βελτιστοποίηση ερωτημάτων
- Υπολογισμός νέφους
- Distributed query optimization
- Adaptive query optimization
- Multiobjective query optimization
- Cloud computing
- Φυσικές Επιστήμες
- Επιστήμη Ηλεκτρονικών Υπολογιστών και Πληροφορική
- Natural Sciences
- Computer and Information Sciences
Rights
- Language dc:language
- gre
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/36711