Comparing Virtualization Solutions for NFV Deployment: a Network ...

4 downloads 345 Views 868KB Size Report
Comparing Virtualization Solutions for. NFV Deployment: a Network Management. Perspective. Lucas Bondan*, Carlos R. P. dos Santos, Lisandro Z. Granville.
Comparing Virtualization Solutions for NFV Deployment: a Network Management Perspective Lucas Bondan*, Carlos R. P. dos Santos, Lisandro Z. Granville *PhD. Student at Federal University of Rio Grande do Sul (UFRGS) - Presenter

21th International Symposium on Computer and Communications (IEEE ISCC 2016) 27 – 30 June, 2016 – Messina, Italy

Outline ● ● ● ● ● ●

Introduction Background & Related Work Methodology Evaluation Results Results Discussion Final Remarks 2

Outline ● ● ● ● ● ●

Introduction Background & Related Work Methodology Evaluation Results Results Discussion Final Remarks 3

1. Introduction

Network Functions Virtualization ● “Softwarize” network functions into Virtual Network Functions (VNF)s [1] ○ Benefits: ○ Challenges: ■ Reduce CAPEX and OPEX ■ Virtualization performance ■ Business and service agility ■ Dynamic VNF migration ■ Quickly generate revenue ■ Service Function Chaining (SFC) ■ Among others ■ Among others, such as... 4

1. Introduction

Management and Orchestration ● First step: select a virtualization solution ○ Direct implications on the network performance ○ Provide management support ○ Satisfy the needs of network operators

● But, what is the best virtualization solution for NFV deployments? ○ From the management point-of-view! 5

Outline ● ● ● ● ● ●

Introduction Background & Related Work Methodology Evaluation Results Results Discussion Final Remarks 6

2. Background & Related Work

Context ● A virtualization solution: ○ Must provide essential virtualization properties to run VNFs Must fulfill network operators needs

○ ● Available solutions are based on different virtualization approaches 7

2. Background & Related Work

Virtualization Solutions ● ClickOS [2] ○ Full virtualization (Xen hypervisor) ○ Optimized for fast network packet processing

● CoreOS [3] ○ Container-based virtualization (Docker) ○ Deployment of wide ranges of isolated functions

● OSv [4] ○ Full virtualization (KVM hypervisor) ○ Able to run functions designed in different languages 8

2. Background & Related Work

Virtualization Solutions ● Main question: How to properly manage NFV deployments composed by these solutions? 9

2. Background & Related Work

Literature ● ETSI NFV MANO [5] ○ Defines some reference points ○ Information model for operational NFV elements ■ Virtual links, VNFs, Virtual Machines (VM)s, … ○ Problem: too vague to be implemented in practice

● IRTF NFV Research Group (NFVRG) ○ Internet-Drafts ■ Policy-based management, service verification, resource management, … ○ Problem: lack of practical analysis 10

2. Background & Related Work

Literature (cont.) ● Previous work: identification of management requirements [6] ○ Difficulties faced by network operators ○ Based on a NFV deployment ○ Using one virtualization solution - ClickOS

● Now: evaluate different solutions ■ Considering the management requirements ■ How? 11

Outline ● ● ● ● ● ●

Introduction Background & Related Work Methodology Evaluation Results Results Discussion Final Remarks 12

3. Methodology

Classification ● Management requirements ○ Qualitative: subjective analysis ○ Quantitative: objective analysis ■ Able to be evaluated using performance metrics

● Quantitative management requirements ○ ○ ○ ○

VMs/containers Instantiation VNF Deployment VM/container & VNF Monitoring Physical and Virtual NFs Coexistence 13

3. Methodology

Performance Metrics ● Boot time ○ Time needed to for a VNF to get ready/running

● Request time ○ Time spent by virtualization solutions to receive, process, and response packets

● Memory consumption ○ Total amount of memory available after VMs/containers instantiation 14

3. Methodology

Evaluation Scenario

15

Outline ● ● ● ● ● ●

Introduction Background & Related Work Methodology Evaluation Results Results Discussion Final Remarks 16

4. Evaluation Results

Boot Time ● Evaluation: ○ Script measures the time needed by a VM or container to start its operation and response to a network request ○ VNF “Ponger” implemented to reply the network request in each VM

17

4. Evaluation Results

Boot Time

18

4. Evaluation Results

Response Time ● Evaluation: ○ “Proxy” implemented in each virtualization solution ■ Forwards packets from A to B and the opposite ○ Very simple and lightweight

19

4. Evaluation Results

Response Time

20

4. Evaluation Results

Memory Consumption ● Evaluation: ○ Memory measured in the VNF server, considering the total amount available ○ Same “Ponger” used in the boot evaluation was instantiated inside the VMs/containers 21

4. Evaluation Results

Memory Consumption

22

Outline ● ● ● ● ● ●

Introduction Background & Related Work Methodology Evaluation Results Results Discussion Final Remarks 23

4. Results Discussion

Management Requirements ● VMs/containers Instantiation ○ Metrics: Boot time and Memory consumption ○ Indicated solution: CoreOS ○ Alternative: OSv - wide range of images

● VNF Deployment ○ Metrics: Boot time and Memory consumption ○ Indicated solutions: ClickOS for dynamic scenarios and CoreOS for static scenarios 24

4. Results Discussion

Management Requirements ● VM/container & VNF Monitoring ○ Metric: Response time ○ Indicated solution: ClickOS

● Physical and Virtual NFs Coexistence ○ Metrics: Response time and Memory consumption ○ Indicated solution: ClickOS 25

Outline ● ● ● ● ● ●

Introduction Background & Related Work Methodology Evaluation Results Results Discussion Final Remarks 26

6. Final Remarks

Conclusions ● ClickOS and CoreOS are superior to OSv ○ Boot time: ClickOS 18,61% faster than CoreOS in the best case ○ Response time: ClickOS 10 times faster than CoreOS for packets 64 KB sized ○ Memory consumption: CoreOS requires 42,34% less memory than ClickOS for 128 instances

● However, MANO solutions should also consider the drawbacks using one or another solution ○ OSv could be a promising alternative for dynamic scenarios 27

6. Final Remarks

Future Work ● Design a comprehensive management system for NFV ○ Covering as many as possible the management requirements

● Investigate orchestration mechanisms for NFV ○ Synergy among ETSI recommendations and our management requirements ○ Unified design pattern 28 for NFV MANO

References [1] M. Chiosi et al., “Network Functions Virtualisation (NFV),” ETSI NFV ISG, White Paper, 2012. Available at: https: //portal.etsi.org/nfv/nfv white paper2.pdf. [2] J. Martins, M. Ahmed, C. Raiciu, V. Olteanu, and M. Honda, “ClickOS and the Art of Network Function Virtualization,” USENIX Symposium on Networked Systems Design and Implementation (NSDI), 2014. [3] CoreOS: Open Source Projects for Linux Containers. Available at: https://coreos.com/. Accessed December, 2015. [4] A. Kivity, D. Laor, G. Costa, P. Enberg, N. Har’El, D. Marti, and V. Zolotarov, “OSv—Optimizing the Operating System for Virtual Machines,” in USENIX Annual Technical Conference (USENIX ATC 14). Philadelphia, PA: USENIX Association, Jun. 2014, pp. 61–72. [5] J. Quittek et al., “Network Functions Virtualisation (NFV) - Management and Orchestration,” ETSI NFV ISG, White Paper, 2014. [6] L. Bondan, C. R. P. d. Santos, and L. Z. Granville, “Management requirements for ClickOS-based Network Function Virtualization,” in International Workshop on Management of SDN and NFV Systems (ManSDN/NFV) collocated with the International Conference on Network and Service Management (CNSM), Nov 2014, pp. 447–450.

29

Thank you! (Grazie!) ● Questions?

Lucas Bondan [email protected] inf.ufrgs.br/~lbondan

30

Proxy algorithm

31

Problem Statement ● No investigations were conducted to show how effective virtualization solutions are with regard to management requirements ● Analysis of different virtualization solutions ○ Helps network operators interested in adopting NFV on their environments. 32