Service-Aware Virtual Network Function Migration Based on Deep Reinforcement Learning

Network Function Virtualization (NFV) aims to provide a way to build agile and flexible networks by building a new paradigm of provisioning network services where network functions are virtualized as Virtual Network Functions (VNFs). Network services are implemented by service function chains, which are formed by a series of VNFs with a specific traversal order. VNF migration is a critical procedure to reconfigure VNFs for providing better network services. However, the migration of VNFs for dynamic service requests is a key challenge. Most VNF migration works mainly focused on static threshold trigger mechanism which will cause frequent migration. Therefore, we propose a novel mechanism to solve the issue in this paper. With the objective of minimizing migration overhead, a stochastic optimization problem based on Markov decision process is formulated. Moreover, we prove the NP-hardness of the problem and propose a service-aware VNF migration scheme based on deep reinforcement learning. Extensive simulations are conducted that the proposed scheme can effectively avoid frequent migration and reduce the migration overhead.