Towards Traffic-Driven VNF Scaling: A Preliminary Case Study based on Container
In Network Function Virtualization (NFV), Virtualized Network Function (VNF) scaling is one of the key lifecycle management operations to accommodate the traffic fluctuation. Compared with a reactive scaling approach based on the load threshold, proactive traffic load prediction can drive the VNF scaling ahead of time and avoid VNF states movement by only redirecting new coming flows. However, most existing online learning research is based on presumed VNF capacity or utilizes a server cluster with high cost and heavy foot-print. To provide the environment for online learning-based VNF scaling research, based on Docker container, we build a lightweight platform on a general personal computer (PC), which supports real word traffic replay and fine-grained resource allocation. Our preliminary case study evaluates the capacity of Snort-based IDS with one CPU core and a half of CPU core under different traffic replay speeds. The experiment results verify that the CPU consummation level rises with the increase of replay speed and the overhead causes packet loss and missing alerts of threads. Besides, under the same replay speed, the CPU consummation level fluctuates with traffic condition. The preliminary case study demonstrates that the container-based platform can provide the basis for online traffic-driven VNF scaling research.