Toward Deploying Parallelized Service Function Chains Under Dynamic Resource Request in Multi-Access Edge Computing

Abstract

Resource distribution policy and how to assemble the Service Function Chain (SFC) in Multi-access Edge Computing (MEC) networks to meet service quality standards poses an important challenge for Network Function Virtualization (NFV) technology. Increasing the number of Virtual Network Functions (VNFs) leads to high-latency SFC assembly, which can be countered by network function parallelization. However, existing studies parallelize VNF for resource allocation in MEC by assuming that the demanded resources do not change during SFC assembly. To address these issues, this paper develops a Latency-aware VNF Parallelization strategy under Resource demand Uncertainty (LVPRU) in MEC. We formulate LVPRU under the assumption of resource uncertainty in MEC via Quadratic Integer Programming (QIP) and show that the problem is NP-hard. LVPRU parallelizes VNFs by discovering dependencies between them and assembles multiple sub-SFCs instead of the original SFC. We apply Asynchronous Advantage Actor-Critic (A3C) as a deep reinforcement learning algorithm to assemble sub-SFCs. We finally evaluate the performance of LVPRU through trace-driven simulations. The evaluation results of proposed strategy are promising in different scenarios compared to benchmark algorithms.