Economic Considerations for Integrating Massively Parallel Heterogeneous Devices into the Cloud

Abstract

Cloud computing providers are expected to supply high-performance and scalable computing resources to their users on demand. Powering these computing resources to provide this service comes with a high operating cost. As the use of cloud computing has become considerably widespread, the workloads provided and computational demands placed upon cloud computing infrastructures have become extremely diverse. Accompanying this, the computational resources available for cloud computing infrastructures are also growing in diversity, yielding heterogeneous computing systems. Through the adoption of heterogeneous computing devices, such as the Graphics Processing Unit, cloud computing providers have a potential avenue for supplying enhanced computational performance with reduced power consumption and in turn, reduced operating cost. In this paper, we show that enhancing heterogeneity through the incorporation of multiple computational accelerators increases performance while also decreasing energy consumption. The benefits provided by heterogeneous computational accelerators leads to improved service rates and operating costs for cloud computing centers.