Performance Impacts of Scaling Policies for virtual IP Multimedia Subsystem on the Cloud

In network function virtualization, mobile network functions are segregated from proprietary hardware to the commercial off-the-shelf computer on the cloud. This paradigm shift aims at reducing capital expenditures while improving the scalability of Virtual Network Functions (VNFs). Horizontal scaling and vertical scaling are two typical scaling policies. This paper investigates the performance impacts while adopting different types of scaling policies into a practical cloud system. A virtual IP Multimedia Subsystem (vIMS) is deployed on an OpenStack testbed for validating the VNF on-boarding. To evaluate the scaling performance of vIMS, the successful call rate is measured with a variety of resource capacities. Experimental results show that the performance of vertical scaling policy has a limitation when the instance size is getting larger. As an alternative, the horizontal scaling policy enhances the performance while increasing the number of instances. Based on the evaluations, a novel auto-scaling algorithm is further proposed to adapt the utilization of VNF instances. Results also verify the feasibility and effectiveness of the proposed algorithm with varied call rates.