National Technical University of Athens (NTUA)
Τεχνολογίες και μηχανισμοί μοντελοποίησης και πρόβλεψης απόδοσης υπηρεσιοστρεφών εφαρμογών και υποδομών
Abstract
dc:descriptionThe advent of service oriented infrastructures providing software and hardware as a service has rendered feasible the utilization of the latter by external users in the form of pay-per-use resources. In order to ensure the demanded quality of service during the provisioning of these resources, a multilayer translation and modeling mechanism is necessary, in order to convert the application terms that are offered in traditional service level agreement contracts to resource level attributes. In the current thesis different candidate methodologies are analyzed with regard to their ability to dynamically create application models in service oriented infrastructures and the fittest one is selected based on the limitations and requirements. Its weaknesses are also investigated and innovative approaches are applied in order to render it applicable to the desired framework. The final form of this method (genetically optimized artificial neural networks) is validated in a variety of real world applications for predicting their performance based on the given hardware of execution, their high level parameters and configuration. What is more, a two level mechanism (workload forecasting and translation) extends its use in cases where the application workload cannot be foreseen by the owner of the service. Through the proposed framework, historical data are analyzed and patterns of usage are discovered that aid in the full automation of the management framework. Furthermore, a multi-layer, decoupled and service oriented framework, based on specialized numerical software, is designed and implemented for incorporating the selected method in cloud computing environments. In addition, the thesis analyzes the performance interference between concurrently running virtual resources and applications in a multitenant service oriented execution framework like Clouds. A limited number of benchmarks is chosen that depicts characteristic usage of the hardware resources and the degradation of their performance is measured due to the coexistence in the same physical host. Parameters that are taken under consideration in this process include also real time scheduling configuration, necessary for guaranteeing a process’s time on the CPU, along with assignment patterns on multicore architectures. Suitable prediction models are created, based on which the IaaS provider may have a priori knowledge of the overhead inserted by the execution of a specific task combination on a physical host and choose the optimal combinations for the guarantee of the actual Quality of Service offered by the resources.
Degree
thesis:*- Grantor dc:publisher
- National Technical University of Athens (NTUA)
- Year dc:date
- 2012
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Kousiouris, Georgios
- Κουσιουρής, Γεώργιος
Subjects
dc:subject × 18- Υπηρεσιοστρεφείς υποδομές
- Υπολογιστικό νέφος
- Τεχνητή νοημοσύνη
- Γενετικοί αλγόριθμοι
- Τεχνητά νευρωνικά δίκτυα
- Πρόβλεψη απόδοσης
- Εικονικοποίηση
- Service oriented infrastructures
- Cloud computing
- Artificial intelligence
- Genetic algorithms
- Artificial neural networks
- Performance prediction
- Virtualization
- Engineering and Technology
- Electrical Engineering, Electronic Engineering, Information Engineering
- Επιστήμες Μηχανικού και Τεχνολογία
- Επιστήμη Ηλεκτρολόγου Μηχανικού, Ηλεκτρονικού Μηχανικού, Μηχανικού Η/Υ
Rights
- Language dc:language
- gre
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/27132