Energy models for NFV and service provisioning on fog nodes

Abstract

The demand for high-bandwidth, low-latency services is rapidly increasing. Content Distribution Networks (CDNs) have addressed this by providing content from within or close to the Internet Service Provider (ISP). Still, the most common bottleneck for high service quality is the 'last mile' between ISP and end user. Serving content from small caches on end-user devices promises to increase service quality of the respective content. Similarly, fog computing promises to provide low-latency services from arbitrary nodes within the network. Both require additional functionality provided by network functions virtualization (NFV), redirecting traffic to the appropriate destinations. Still, the cost and performance of possible solutions are not well analyzed. Hence, this paper analyzes the forwarding and computing performance of a number of single-board computers (SBCs) from which models for the performance and energy cost of different loads are derived. Furthermore, the development of energy efficiency gains over the last years is analyzed, confirming Koomey's law also for SBCs, leading to an increase of computational efficiency by a factor of 5.5 to 7.5 over the course of four years, which relates to a doubling time of 1.62 years.