About the Post

Author Information

Michele Mazzucco is a Post-Doctoral Research Fellow in the Software Engineering Group at the University of Tartu (Estonia) and a Researcher at the Software Technology and Applications Competence Center (STACC).

Maximizing Cloud Providers Revenues via Energy Aware Allocation Policies

Cloud providers, like Amazon, offer their data centers’ computational and storage capacities for lease to paying customers. High electricity consumption, associated with running a data center, not only reflects on its carbon footprint, but also increases the costs of running the data center itself.

In “Maximizing Cloud Providers Revenues via Energy Aware Allocation Policies” we address the problem of maximizing the revenues of Cloud providers by trimming down their electricity costs. As a solution we introduce and evaluate allocation policies which are based on the dynamic powering servers on and off. The policies aim at satisfying the conflicting goals of maximizing the users’ experience while minimizing the amount of consumed electricity. The results of several experiments are described, showing that the proposed scheme performs well under different traffic conditions. In particular, we find that decisions such as how many servers are powered on can have a significant effect on the revenue earned by the provider. Moreover, those decisions are affected by the contractual obligations between clients and provider.

We have extended this work by developing allocation policies resilient to errors in the forecasting, as well as a method for finding the allocation policies parameters leading to the highest revenues. In the figure below we compare the performance achieved by the heuristic policies we have designed to that obtained by the `Oracle’ policy, i.e., a policy which knows what the future user demand will be. The experiment simulated one month of traffic using a scaled version of the Wikipedia traces. As we can see, the ‘Grassmann’s’ heuristic demonstrates the results almost identical to the best possible in terms of revenue, while ‘QED’ and ‘Predictive’ still demonstrate very strong results. The most interesting property of the ‘Grassmann’s’ heuristic is that even though it entails slightly higher power consumption, it results in a lower number of lost jobs, which positively reflects on the reputation of the Cloud provider in the long term. The ‘Optimal’ policy, instead, tries to maximize the revenue only.

N.B. The data about the performance of the policies specifically designed to deal with non-stationary traffic is not publicly available yet. However feel free to contact me for more information.

Maximizing Cloud Providers Revenues via Energy Aware Allocation Policies

Performance of the heuristics compared to that obtained by the Oracle policy.

No comments yet.

Leave a Reply

Connect with: