Efficient bare metal auto-scaling for NFV in edge computing

Abstract

Elasticity is an essential attribute of cloud data center, which is critical for operating resources in face of peaks and valleys of business. At present, the automatic scaling technique of virtual machines is widely studied, but barely for physical machines. Despite lack of flexibility, we all know that physical server can perform faster and more efficiently than virtualized instances, especially in Network Function Virtualization (NFV) systems. Some virtual network functions (VNFs) actually require high performance computing, which is a hard task for virtual machines. Besides, good management of bare metal resources can be significant for the data center power cost and human maintenance cost. Accordingly, we think that auto-scaling of physical machine is worth studying. This paper proposes a bare metal automatic scaling scheme based on workload prediction, and finally make tests on an open source NFV platform. The new scheme obtains good result on computation intensive VNFs scenario, including complete the scale in minutes, guarantee for the continuity of VNF processing business, and can cope with the load fluctuation better.