近年來5G 網路的蓬勃發展，以及網路切片（Network Slicing）的概念出現，網路功能虛擬化（Network Function Virtualization, NFV）吸引了大量的研究關注。運營商可以在運營的服務器上，佈建並使用各式各樣的網路功能（Network Function），例如：網路監視器，有狀態防火牆，入侵檢測系統，相較於使用傳統的中間箱（Middlebox），需要根據不同的服務購買不同的網路功能，前者節省了成本以及增加了彈性。另外，隨著最近發布的混合式多核心架構，例如: 蘋果M1 處理器、英特爾12 代晶片，使得如何在不浪費資源的情況下在混合式多核處理器間分配CPU 資源成為一個重要議題。
在本篇論文中，我們提出了一種分配演算法，該算法決定如何在混合式多核心架構上部署網絡功能，以實現審慎性和負載平衡。為了做到審慎性，所有（除了一個）處於喚醒狀態的核心的使用率必須超過最低閾值； 也就是說，最多只有一個清醒核心未被充分利用。對於負載平衡，我們更偏好盡可能多的清醒、充分利用的核心；這最大限度地提高了計算能力，並最大限度地減少了隨著流量變化而喚醒額外核心的可能性。基於預測CPU 核心的使用率，我們提出的算法利用稱為cgroups 的Linux 內核功能來限制、負責和隔離進程的運行。此外，我們的方法能夠適應動態流量變化，並且能夠在必要時在CPU 內核之間遷移網絡功能。在性能評估中，我們提出的方法優於其他方法。

With the recent boom of 5G networks and the emergence of the concept of network slicing, Network Function Virtualization (NFV) has attracted a lot of research attention. Operators can deploy a variety of network functions, such as network monitors, stateful firewalls, and intrusion detection systems, on commodity servers. Compared to traditional middleboxes, different network functions are purchased for different services, thus saving costs and increasing flexibility. In
addition, the recent release of hybrid multi-core architecture, such as Apple’s M1 and Intel’s 12th-generation Core processors, makes how to allocate CPU resources among multi-core processors without wasting resources an important issue.
In this thesis, we propose an assignment algorithm that decides how to deploy network functions on hybrid multi-core architecture for prudence and load balancing. For prudence, the usage percentages of all (except one) cores that are awake must exceed a minimum threshold; that is, at most one of the awake cores is underutilized. For load balancing, we prefer as many awake, non-underutilized cores as possible; this maximizes computation capacity and minimizes the probability of awakening additional cores as traffic varies. Based on predicted CPU core usage, the algorithm we propose utilizes the Linux kernel feature called cgroups to limit, account for, and isolate the running of processes. In addition, our method is adaptive to dynamic traffic changes and is able to migrate network functions between CPU cores if necessary. In performance evaluation, our proposed method
outperforms other methods.

Abstract.................................................. i
中文摘要.................................................. iii
Contents.................................................. iv
List of Figures........................................... vi
1 Introduction............................................ 1
2 Related Work............................................ 5
3 System Model............................................ 8
3.1 Problem Statement......................................8
3.1.1 Prudence and Load balancing......................... 9
3.1.2 Avoidance of Contention of NFs from the Same SFC.... 11
3.1.3 Dynamic Quota Adjustment and Task Migration......... 12
3.2 Processing Flow....................................... 14
4 Proposed Method......................................... 15
4.1 MTCF for Prudence and Load Balancing...................15
4.2 NF-exchange for Avoidance of Contention of NFs from the Same SFC ...........................................................24
4.3 NF Migration Mechanism for Dynamic Quota Adjustment and Task
Migration................................................. 26
5 Implementation ......................................... 28
5.1 Experimental Environment.............................. 28
5.1.1 NFV Framework ...................................... 28
5.1.2 Packet Generator ................................... 29
5.1.3 Hardware ........................................... 29
5.2 Implementation Details................................ 31
5.2.1 Cgroups-based Process Isolation .................... 31
5.2.2 Provisioning of Multiple Service Function Chains on Open-
NetVM..................................................... 33
5.2.3 Machine-learning (ML) CPU Usage Prediction.......... 34
5.3 Performance Measurement............................... 35
6 Evaluation.............................................. 37
6.1 Cgroups-based NFs Isolation........................... 38
6.2 Dynamic Migration Mechanism........................... 39
6.3 Simulation............................................ 41
6.3.1 Load Balancing...................................... 43
6.3.2 CPU Usage........................................... 44
6.3.3 SFC Repulsion....................................... 46
7 Conclusion.............................................. 49
Reference................................................. 50

[1] ETSI, Network Functions Virtualisation (NFV); Architectural Framework, https://www.etsi.org/deliver/etsi_gs/nfv/001_099/002/01.02.01_60/gs_nfv002v010201p.pdf, [Online; accessed 1-June-2022], 2014.
[2] W. Zhang, J. Hwang, S. Rajagopalan, K. Ramakrishnan, and T. Wood,“Flurries: Countless fine-grained nfs for flexible per-flow customization,” in Proceedings of the 12th International on Conference on Emerging Networking EXperiments and Technologies, 2016, pp. 3–17.
[3] C. Sun, J. Bi, Z. Zheng, H. Yu, and H. Hu, “Nfp: Enabling network function parallelism in nfv,” in Proceedings of the ACM SIGCOMM Conference, Aug. 2017, pp. 43–56.
[4] A. Chowdhary and D. Huang, “Sdn based network function parallelism in cloud,” in 2019 International Conference on Computing, Networking and Communications (ICNC), 2019, pp. 486–490.
[5] Z. Ni, G. Liu, D. Afanasev, T. Wood, and J. Hwang, “Advancing network function virtualization platforms with programmable nics,” in 2019 IEEE International Symposium on Local and Metropolitan Area Networks (LANMAN), 2019, pp. 1–6.
[6] B. Deng, W. Wu, and L. Song, “Redundant logic elimination in network functions,” in Proceedings of the Symposium on SDN Research, 2020, pp. 34–40.
[7] A. Mohammadkhan, K. K. Ramakrishnan, and V. A. Jain, “Cleang—improving the architecture and protocols for future cellular networks with nfv,” IEEE/ACM Transactions on Networking, no. 6, pp. 2559–2572, 2020.
[8] S. G. Kulkarni, W. Zhang, J. Hwang, et al., “Nfvnice: Dynamic backpressure and scheduling for nfv service chains,” IEEE/ACM Transactions on Networking, vol. 28, no. 2, pp. 639–652, Feb. 2020.
[9] M. Ghaznavi, E. Jalalpour, B. Wong, R. Boutaba, and A. J. Mashtizadeh, “Fault tolerant service function chaining,” in Proceedings of the Annual conference of the ACM Special Interest Group on Data Communication (SIGCOMM) on the applications, technologies, architectures, and protocols for computer communication, Jul. 2020, pp. 198–210.
[10] M. Reininger, A. Arora, S. Herwig, et al., “Bento: Safely bringing network function virtualization to tor,” in Proceedings of the 2021 ACM SIGCOMM 2021 Conference, 2021, pp. 821–835.
[11] J. Gil Herrera and J. F. Botero, “Resource allocation in nfv: A comprehensive survey,” IEEE Transactions on Network and Service Management, pp. 518–532, 2016.
[12] J. Li, W. Shi, Q. Ye, W. Zhuang, X. Shen, and X. Li, “Online joint vnf chain composition and embedding for 5g networks,” in 2018 IEEE Global Communications Conference (GLOBECOM), 2018, pp. 1–6.
[13] M. Savi, M. Tornatore, and G. Verticale, “Impact of processing-resource sharing on the placement of chained virtual network functions,” IEEE Transactions on Cloud Computing, pp. 1479–1492, 2021.
[14] P. Kr¨amer, P. Diederich, C. Kr¨amer, R. Pries, W. Kellerer, and A. Blenk, “Sfc2cpu: Operating a service function chain platform with neural combinatorial optimization,” in 2021 IFIP/IEEE International Symposium on Integrated Network Management (IM), 2021.
[15] AWS, Amazon EC2, https://aws.amazon.com/ec2/?nc1=h_ls, [Online;
accessed 29-July-2022], 2022.
[16] Z. Wei, How to use DPDK in AWS EC2 instances and AWS-based container platforms, https://aws.amazon.com/cn/blogs/china/how- to- usedpdk-in-aws-ec2-instances-and-aws-based-container-platformsi/,
[Online; accessed 29-July-2022], 2019.
[17] J. Csirik, “Two simple algorithms for bin covering,” Acta Cybernetica, vol. 14, no. 1, pp. 13–25, Jan. 1999.
[18] C.-F. Kuo, “Nfv performance prediction based on run-time instruction analysis,” M.S. thesis, National Tsing Hua University, Jul. 2021.
[19] W. Zhang, G. Liu, W. Zhang, et al., “OpenNetVM: A Platform for High Performance Network Service Chains,” in Proceedings of the ACM SIGCOMM Workshop on Hot Topics in Middleboxes and Network Function Virtualization, Aug. 2016.
[20] K. Wiles, The Pktgen Application, https://pktgen-dpdk.readthedocs. io/en/latest, [Online; accessed 11-July-2022], 2010.
[21] Intel, Intel Core i9-12900K Processor, https://www.intel.com/content/www/us/en/products/sku/134599/intel-core-i912900k-processor-30m-cache-up-to-5-20-ghz/specifications.html, [Online; accessed 1-June-2022], 2021.
[22] W1zzard, Intel Core i9-12900K E-Cores Only Performance Review, https://www.techpowerup.com/review/intel- core- i9- 12900k- e- coresonly-performance/2.html, [Online; accessed 1-June-2022], 2021.
[23] J. Zheng, C. Tian, H. Dai, et al., “Optimizing nfv chain deployment in software-defined cellular core,” IEEE Journal on Selected Areas in Communications, no. 2, pp. 248–262, 2020.