AXaaS: Case for Acceleration as a Service

Abstract

The ubiquity and the range of utility of ”smart” devices is ever increasing. Device limitations have lead developers to leverage cloud-offloading to gain performance for their applications. As users become aware of the expanding utility of their devices through these powerful applications, they tend to demand more from them. However, developers’ intent on providing state-of-the-art applications will undoubtedly hit performance barriers for emerging products due to the inherently high latency of the prevailing mobile-cloud architecture. This paper proposes a new type of service architecture called AXaaS (Acceleration as a Service) that will empower developers to satisfy user demand for greater application performance and fully realize new computationally-intensive applications that would be otherwise impossible or impractical. While Telecom Service Providers (TSP) already provide data and bandwidth services, we introduce a new paradigm in which the TSP may charge subscribers for computational acceleration of complex applications by outsourcing computational tasks to larger cloud operators. We provide an exposition of the performance potential of such a service by examining its theoretical impact upon an opensource- based Face Recognition application. We also examine a sample instantiation of cloud resources via AmazonWeb Services, and estimate the return on investment for a TSP implementing AXaaS. We find the TSP-side ROI to be quite favorable, which means that AXaaS is a viable new aaS alternative. I