[Back]

G. Copil:
"Cloud Services Elasticity Control: from requirements specification to operations management";
Supervisor, Reviewer: S. Dustdar, H. Truong, N. Koziris; Institut für Informationssysteme, Distributed Systems Group, 2016; oral examination: 2016-02-12.

Utility based computing, such as cloud computing, is a driving force for the adoption of highly distributed applications (e.g., microservices-based applications). Through the use of cloud computing, application stakeholders can choose from a multitude of
configurations for their application deployment, e.g., public-private deployments, multicloud deployments, and can adapt to whatever is best for their application from the technology and business perspectives. After the application deployment, stakeholders may
choose to offer the application as a cloud service (i.e., Software as a Service), benefiting from cloud providers services (e.g., resources on-demand offered by Infrastructure as a Service providers or platforms on-demand offered by Platform as a Service providers). For
fully using cloud-offered services, the application needs to be adapted by its stakeholders, considering the load, and their users´ requirements, to obtain the desired quality at the minimum price to be paid to cloud providers in exchange for their resources. However, for achieving this, the application owner needs to pay a specialized person for monitoring and adapting whenever needed the application, or to use existing solutions offered by cloud providers which enable solely simple scaling, without considering complex applications,
application-level configurations, and dependencies among application components.
In this thesis, a framework for elasticity control of cloud services is proposed. Starting from high-level requirements, a language, SYBL, is proposed, which supports service stakeholders in describing their requirements at multiple levels of abstraction. With this language as a basis, a framework, rSYBL, is proposed, in order to control cloud
service elasticity considering its behavior and the specified requirements. For improving the control, the behavior of cloud services, in time, at multiple levels of abstraction, is estimated, based on mechanisms that are also integrated in rSYBL. The framework is further extended for supporting operations management at runtime, considering elasticity concerns, integrating stakeholders/employees in the whole control process. Using an illustrative case study, each of the proposed mechanisms is evaluated. Results show that rSYBL is able to control cloud services, during runtime using monitoring information and requirements coming from stakeholders.