近年來網路功能虛擬化（Network Function Virtualization, NFV）逐漸受到網路服務供應商重視。相較於傳統專用的網通設備一個裝置只能提供一項功能，此乃受限於硬體、電路設計；網路功能虛擬化的願景是使用唾手可得之通用型電腦主機，運行所需要的網路服務如：防火牆、入侵偵測系統、代理伺服器等等，不受限於專用硬體即可增加資源利用彈性，也能降低被特定網通設備廠商綁架的風險。

然而採用通用型電腦主機缺點在於效能不易掌控，此乃該主機非專為網通用途設計，容易受到：系統中斷（interrupt）、快取未命中（cache miss）、資源競爭等影響。為了確保網路使用不會中斷與最大化資源效益，效能預測便成為重要的議題。

在本篇論文中會針對現有常見的虛擬網路設備（Virtual Network Function, VNF）進行執行階段分析、蒐集必要效能影響因子，透過機器學習方式訓練並預測不同設定檔、NF、網路流量下的效能表現。

Nowadays, Network Function Virtualization (NFV) is flourishing among network service providers. Compared to traditional network equipment which provides only one function per device (due to the hardware limitation and dedicated circuit design), NFV which uses the COTS (commercial-off-the-shelf) servers instead of dedicated hardware to provide network services like firewall, IDS, proxy, etc. can provide more flexibility.

However, one drawback of COTS servers is unmanageable performance; the reason is that they are not designed for network processing purposes and susceptible to interrupt, cache miss, and resource contention. To ensure network availability and maximize resource efficiency, performance prediction has been an important issue.

In this manuscript, we analyze and collect the performance metrics of Virtual Network Function (VNF) during run time. With the help of machine learning, we can predict the performance under different configurations of VNFs and at different packet rates.

Abstract........................................ i
中文摘要........................................ ii
誌謝辭.......................................... iii
Contents........................................ iv
List of Figures................................. vii
1 Introduction.................................. 1
2 NFV Technology................................ 3
2.1 Recent Research............................. 3
2.2 Architecture................................ 3
2.3 Poll Mode vs Interrupt Mode ................ 6
2.4 Batch Processing ........................... 6
3 Related Work.................................. 7
3.1 Performance Enhancement..................... 7
3.2 Verification and Diagnosis ................. 8
3.3 High Availability .......................... 9
3.4 Performance Prediction...................... 9
3.5 Design Challenges........................... 10
4 System Model.................................. 11
4.1 Problem Statement........................... 11
4.2 Key Idea.................................... 12
4.3 Processing Flow............................. 13
5 Proposed Method............................... 14
5.1 Collect Metrics ............................ 14
5.1.1 Performance Tool Selection ............... 14
5.1.2 Data Collecting Approach ................. 15
5.2 Train Model................................. 17
5.2.1 Instruction Feature Selection............. 17
5.2.2 Machine Learning Method................... 18
6 Implementation................................ 19
6.1 Environment................................. 19
6.1.1 MANO...................................... 19
6.1.2 Packet.................................... 20
6.1.3 Hardware ................................. 20
6.1.4 MISC ..................................... 21
6.2 Details..................................... 23
6.2.1 Improve Unstable Batch Size .............. 23
6.2.2 Collect Instruction Metrics............... 27
6.2.3 Increase NF Variety....................... 30
6.2.4 NF Topology .............................. 31
6.2.5 Collect Data.............................. 33
6.3 Preprocess & Analysis ...................... 34
7 Experimental Evaluation....................... 37
7.1 Raw Data of UDP Packet ..................... 38
7.2 Prediction Result of UDP Packet ............ 39
7.3 Analysis.................................... 43
7.4 Additional TCP Evaluation .................. 44
8 Conclusion and Future Work.................... 47
Reference....................................... 49

[1] R. Cziva and D. P. Pezaros, “Container network functions: Bringing nfv to the network edge,” IEEE Communications Magazine, vol. 55, no. 6, pp. 24– 31, 2017.
[2] G. Liu, Y. Ren, M. Yurchenko, K. Ramakrishnan, and T. Wood, “Mi- croboxes: High performance nfv with customizable, asynchronous tcp stacks and dynamic subscriptions,” in Proceedings of the ACM SIGCOMM Con- ference, Aug. 2018, pp. 504–517.
[3] S. Han, K. Jang, A. Panda, S. Palkar, D. Han, and S. Ratnasamy, “Soft- nic: A software nic to augment hardware,” EECS Department, University of California, Berkeley, Tech. Rep. UCB/EECS-2015-155, 2015.
[4] W. Zhang, G. Liu, W. Zhang, N. Shah, P. Lopreiato, G. Todeschi, K. Ra- makrishnan, and T. Wood, “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.
[5] S. G. Kulkarni, W. Zhang, J. Hwang, S. Rajagopalan, K. Ramakrishnan, T. Wood, M. Arumaithurai, and X. Fu, “Nfvnice: Dynamic backpressure and scheduling for nfv service chains,” IEEE/ACM Transactions on Networking, vol. 28, no. 2, pp. 639–652, Feb. 2020.
[6] E. Kohler, R. Morris, B. Chen, J. Jannotti, and M. F. Kaashoek, “The click modular router,” ACM Transactions on Computer Systems (TOCS), vol. 18, no. 3, pp. 263–297, Aug. 2000.
[7] S. Palkar, C. Lan, S. Han, K. Jang, A. Panda, S. Ratnasamy, L. Rizzo, and S. Shenker, “E2: A framework for nfv applications,” in Proceedings of the Symposium on Operating Systems Principles, Oct. 2015, pp. 121–136.
[8] A. Bremler-Barr, Y. Harchol, and D. Hay, “Openbox: A software-defined framework for developing, deploying, and managing network functions,” in Proceedings of the ACM SIGCOMM Conference, Aug. 2016, pp. 511–524.
[9] 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.
[10] Y. Jiang, Y. Cui, W. Wu, Z. Xu, J. Gu, K. Ramakrishnan, Y. He, and X. Qian, “Speedybox: Low-latency nfv service chains with cross-nf runtime consolidation,” in IEEE International Conference on Distributed Computing Systems (ICDCS), IEEE, Jul. 2019, pp. 68–79.
[11] S. K. Fayaz, T. Yu, Y. Tobioka, S. Chaki, and V. Sekar, “Buzz: Testing context-dependent policies in stateful networks,” in USENIX Symposium on Networked Systems Design and Implementation (NSDI), 2016, pp. 275–289.
[12] A. Fogel, S. Fung, L. Pedrosa, M. Walraed-Sullivan, R. Govindan, R. Maha- jan, and T. Millstein, “A general approach to network configuration analysis,” in USENIX symposium on networked systems design and implementa- tion (NSDI), 2015, pp. 469–483.
[13] N. P. Lopes, N. Bjørner, P. Godefroid, K. Jayaraman, and G. Varghese, “Checking beliefs in dynamic networks,” in USENIX Symposium on Net- worked Systems Design and Implementation (NSDI), 2015, pp. 499–512.
[14] A. Zaostrovnykh, S. Pirelli, L. Pedrosa, K. Argyraki, and G. Candea, “A formally verified nat,” in Proceedings of the ACM SIGCOMM Conference, Aug. 2017, pp. 141–154.
[15] J. Gong, Y. Li, B. Anwer, A. Shaikh, and M. Yu, “Microscope: Queue- based performance diagnosis for network functions,” in Proceedings of the Annual conference of the ACM Special Interest Group on Data Commu- nication (SIGCOMM) on the applications, technologies, architectures, and protocols for computer communication, Jul. 2020, pp. 390–403.
[16] S. G. Kulkarni, G. Liu, K. Ramakrishnan, M. Arumaithurai, T. Wood, and X. Fu, “Reinforce: Achieving efficient failure resiliency for network function virtualization based services,” in Proceedings of the International Conference on emerging Networking EXperiments and Technologies, Dec. 2018, pp. 41– 53.
[17] M. Ghaznavi, E. Jalalpour, B. Wong, R. Boutaba, and A. J. Mashtizadeh, “Fault tolerant service function chaining,” in Proceedings of the Annual con- ference of the ACM Special Interest Group on Data Communication (SIG- COMM) on the applications, technologies, architectures, and protocols for computer communication, Jul. 2020, pp. 198–210.
[18] R. Iyer, L. Pedrosa, A. Zaostrovnykh, S. Pirelli, K. Argyraki, and G. Candea, “Performance contracts for software network functions,” in USENIX Sym- posium on Networked Systems Design and Implementation (NSDI), Boston, MA: USENIX Association, Feb. 2019, pp. 517–530, isbn: 978-1-931971-49-2.
[19] A. Manousis, R. A. Sharma, V. Sekar, and J. Sherry, “Contention-aware performance prediction for virtualized network functions,” in Proceedings of the Annual conference of the ACM Special Interest Group on Data Commu- nication (SIGCOMM) on the applications, technologies, architectures, and protocols for computer communication, Jul. 2020, pp. 270–282.
[20] P. Okelmann, L. Linguaglossa, F. Geyer, P. Emmerich, G. Carle, and T. Paris, “Adaptive batching for fast packet processing in software routers using machine learning,”
[21] R. D. Thomas Willhalm, Intel Performance Counter Monitor, https://github.com/opcm/pcm, [Online; accessed 27-May-2021], 2012.
[22] S. Lange, L. Linguaglossa, S. Geissler, D. Rossi, and T. Zinner, “Discrete- time modeling of nfv accelerators that exploit batched processing,” in IEEE INFOCOM 2019-IEEE Conference on Computer Communications, IEEE, 2019, pp. 64–72.