SLA-Driven Resource Provisioning in the Cloud

Abstract

The common way to purchase computing capacity is to acquire a fix amount of servers which are hosted in a data center. The reserved amount of resources depends on the maximum demand that could occur. This over provisioning of resources is associated with high costs. Cloud providers promise to reduce the costs while guaranteeing always a good performance for the outsourced services. A reducing of the costs can be achieved by a better utilization of the resources for example. To optimize resource utilization while complying to the negotiated Service Level Agreements, the minimum amount of resources has to be determined as exactly as possible. Based on the simulated reprocessing of requests from recent history with different numbers of instances, the optimal number can be found in a short time. Further this work shows that an additional reduction of resources can be achieved by rescheduling the requests before processing.