Joint Virtual Network Function Placement and Flow Routing in Edge-Cloud Continuum

• Published in: IEEE transactions on computers Vol. 73; no. 3; pp. 872 - 886
• Main Authors: Mao, Yingling, Shang, Xiaojun, Liu, Yu, Yang, Yuanyuan
• Format: Journal Article
• Language: English
• Published: New York IEEE 01.03.2024; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Algorithms; Approximation; Approximation algorithms; Cloud computing; Costs; Edge computing; Heuristic algorithms; joint resource and latency optimization; Mathematical analysis; Network function virtualization; Optimization; Routing; Routing (telecommunications); Servers; service function chain deployment; Virtual networks
• ISSN: 0018-9340; 1557-9956
• DOI: 10.1109/TC.2023.3347671

Network Function Virtualization (NFV) is becoming one of the most popular paradigms for providing cost-efficient, flexible, and easily-managed network services by migrating network functions from dedicated hardware to commercial general-purpose servers. Despite the benefits of NFV, it remains a challenge to deploy Service Function Chains (SFCs), placing virtual network functions (VNFs) and routing the corresponding flow between VNFs, in the edge-cloud continuum with the objective of jointly optimizing resource and latency. In this paper, we formulate the SFC Deployment Problem (SFCD). To address this NP-hard problem, we first introduce a constant approximation algorithm for a simplified SFCD limited at the edge, followed by a promotional algorithm for SFCD in the edge-cloud continuum, which also maintains a provable constant approximation ratio. Furthermore, we provide an online algorithm for deploying sequentially-arriving SFCs in the edge-cloud continuum and prove the online algorithm achieves a constant competitive ratio. Extensive simulations demonstrate that on average, the total costs of our offline and online algorithms are around 1.79 and 1.80 times the optimal results, respectively, and significantly smaller than the theoretical bounds. In addition, our proposed algorithms consistently outperform the popular benchmarks, showing the superiority of our algorithms.