MERIT: Model-driven Rehoming for VNF Chains

Abstract

Network service providers often run service chains of Virtual Network Functions (VNFs) on privately owned clouds with limited capacity. These specialized service chains need to meet strict Service Level Objectives (SLOs), especially along the lines of availability (e.g., First responder services). Hence, VNFs in such thinly provisioned clouds may need to be frequently moved, or rehomed, when reacting to various cloud events like hotspots, failures and upgrades. In this paper, we perform a detailed measurement study to show that naive strategies for rehoming, applied uniformly across all VNFs of the service chain, are often sub-optimal when considering different metrics like the user-perceived service downtime and the provider-incurred time delay to complete the rehoming. We propose a novel ModEl-driven RehomIng Technique (MERIT) for VNF chains and empirically analyze the effect of various system parameters on different rehoming actions. Based on our analysis, we develop generic rehoming cost models and further, design and implement an autonomous rehoming system based on MERIT that identifies and executes the optimal rehoming action for each VNF in a service chain. Our experimental results on OpenStack using real-world chains show that MERIT can reduce the chain rehoming delay by up to 47% and the chain downtime by up to 49%.