Diploma and Master Theses (authored and supervised):
"Energie Effizienter Elastizitätsmanager für Clouds";
Supervisor: S. Dustdar, I. Brandic;
Institut für Informationssysteme, AB Verteilte Systeme,
final examination: 2013-11-18.
Elasticity is one of the key features which distinguish clouds from common datacentres. Clouds featuring Elasticity Management react to customer requirements autonomously. Customers can request, use and release resources of the cloud at any point in time. Thus, the amount of resources available for each customer may vary unpredictably.
We designed a predictive approach for autonomic reactions of clouds to the changing requirements of a customerīs application. Our method predicts future resource requirements based on past customer activities. Virtual Machines (VMs) are created as soon as the system detects an indicator for increase in future activities. As soon as the activity drops again, VMs are released to reduce energy costs.
The described method was implemented in the Energy Efficient Elasticity Manager (E3M) to control the elasticity of a customer application and it was evaluated in a simulation. The E3M was compared to other non-predictive elasticity methods in different
test-scenarios, including long-term planning and sudden changes in customer requirements. We show that change in cloud behaviour is unperceivable for customers, as every VM required is available at any point in time. Therefore, our approach helps the cloudprovider
to reduce the consumed energy without reducing the performance of the cloud.
Created from the Publication Database of the Vienna University of Technology.