Date of Award

3-12-2020

Publication Type

Doctoral Thesis

Degree Name

Ph.D.

Department

Computer Science

Keywords

Machine learning, Optimization, QoS, SLA, Web Services

Supervisor

Xiaobu Yuan

Rights

info:eu-repo/semantics/openAccess

Abstract

Quality of service (QoS) is an official agreement that governs the contractual commitments between service providers and consumers in respect to various nonfunctional requirements, such as performance, dependability, and security. While more Web services are available for the construction of software systems based upon service-oriented architecture (SOA), QoS has become a decisive factor for service consumers to choose from service providers who provide similar services. QoS is usually documented on a service-level agreement (SLA) to ensure the functionality and quality of services and to define monetary penalties in case of any violation of the written agreement. Consequently, service providers have a strong interest in keeping their commitments to avoid and reduce the situations that may cause SLA violations.However, there is a noticeable shortage of tools that can be used by service providers to either quantitively evaluate QoS of their services for the predication of SLA violations or actively adjust their design for the avoidance of SLA violations with optimized service reconfigurations. Developed in this dissertation research is an innovative framework that tackles the problem of SLA violations in three separated yet connected phases. For a given SOA system under examination, the framework employs sensitivity analysis in the first phase to identify factors that are influential to system performance, and the impact of influential factors on QoS is then quantitatively measured with a metamodel-based analysis in the second phase. The results of analyses are then used in the third phase to search both globally and locally for optimal solutions via a controlled number of experiments. In addition to technical details, this dissertation includes experiment results to demonstrate that this new approach can help service providers not only predicting SLA violations but also avoiding the unnecessary increase of the operational cost during service optimization.

Share

COinS