Exploiting Price and Performance Tradeoffs in Heterogeneous Clouds

Parallel applications are composed of several tasks, which have different computational demands among them. Moreover, most cloud providers offer multiple instance configurations, with large variations of computational power and cost. A combination between the application requirements and the variety of instance types of the cloud could be explored to improve the cost efficiency of the application execution. In this paper, we introduce the cost-delay product as a metric to measure the cost efficiency of cloud systems. With this metric, cloud tenants can evaluate different tradeoffs between cost and performance for their application, depending on their preferences. We explore the use of multiple instance types to create heterogeneous cluster systems in the cloud. Our results show that heterogeneous clouds can have a better cost efficiency than homogeneous systems, reducing the price of execution while maintaining a similar application performance. Furthermore, by comparing the cost-delay product, the user can select an instance mix that is most suitable for his needs.