Minimizing Resource Rent Loss while Maximizing User Availability in Cloud Applications through Online Switching of the Scaling Method

摘要

Clouds offer access to virtually unlimited resources. Applications requiring a variable number of resources (i.e., elastic) are hence ideal for such a scenario. As cloud resources are usually paid by the hour applications need efficient ways of predicting their requirements in order to reduce costs. In addition we need to maximize user satisfaction through high application availability. Finding the best balance between them is a crucial task. In this paper we consider the case of commercial web applications and focus on the hit rate as scaling factor. Since it is known that no scaling algorithm can have a definite edge over the rest we propose three strategies for online scaling strategy switching. The first one chooses at each step always the best strategy. The second one focuses on analysing the probability distribution of the historic hit rate and determining the best scaling method for it. Finally, the third one relies on changing the method without needing to know the distribution through random choices. Results show that given the fluctuations and uncertainty of cloud environments and user behaviour the most suited method for online selection of the scaling strategy is the first one based on the best strategy.