A Service Level Agreement (SLA) is a documented result of a negotiation between a business customer and a service provider that defines the target values of the Quality of Service (QoS), the responsibilities and rights of every involved party. SLAs become increasingly important in modern business, especially when more and more business applications are deployed in the Software as a service (SaaS) model in cloud computing environments. It is essential for the service provider to prevent SLA violations in order to avoid penalty payments and enhance customer satisfaction. In this work a proactive Service Level Agreement (pSLA) is introduced to improve service performance and prevent SLA violations from two aspects, i.e. proposing reasonable SLA specifications and proactive SLA enforcement. Since different customers’ requests have various requirements, it greatly makes sense for service providers to propose achievable SLA specifications with reasonable costs. The pSLA architecture proposed in this work provides a feedback database as a background knowledge repository, which facilitate SLA negotiation based on the previous experiment data. Furthermore, service providers would like to detect potential SLA violations in advance so that they have the possibility to take counteractions to prevent SLA violations. In pSLA machine learning algorithms are applied to predict QoS target values. If potential violations are detected, pSLA is able to proactively take actions for SLA enforcement. These two features of pSLA benefit from each other so that pSLA can improve the performance of the on-demand services and prevent SLA violations. An experiment based on text analytics and cloud computing is implemented in this work to evaluate the pSLA architecture in the distributed computing environment.
The expected outcome of this master's thesis is to develop a working prototype of a proactive SLA management architecture based on historical performance data and active enforcement at runtime. A scalable text analytics framework is implemented in the cloud as an experiement to evaluate the proposed architecture.
August 2010 - March 2011
There are no subpages or files.