NTT Network Service Systems Laboratories

Management of managed self-organizing network in

network virtualization environment

WTC 2012 @ Miyazaki, Japan

Takashi Miyamura1 Yuichi Ohsita 2 Shohei Kamamura1 Yuki Koizumi2

Shin’ichi Arakawa 2 Kohei Shiomoto1 Atsushi Hiramatsu1 Masayuki Murata 2

1 NTT Network Service Systems Laboratories 2Osaka University

NTT Network Service Systems Labs.

Outline of Talk

•Background

•Problem Statement

•Proposal: Managed self-organization

•Evaluation

•Conclusion

2

NTT Network Service Systems Labs.

Outline of Talk

•Background

•Problem Statement

•Proposal: Managed self-organization

•Evaluation

•Conclusion

3

NTT Network Service Systems Labs.

Overview of Managed Self-Organizing Network

• Aim: Simplify management of network virtualization infrastructure through self-organization

• Challenge: Performance degradation due to resource contention • Proposal: Architecture for stable accommodation self-organizing VNs

4 Optical infrastructure

（physical networks）

PN Resource Manger

VN Agent Controller

VN Agent

VNT derived by Attractor selection

Traffic IN

Traffic OUT

gene1

gene2

gene3

gene4

substrate1

substrate2

substrate4

substrate3

diffusion metabolic network

metabolic reaction

catalyzation

growth rate

diffusion

Bio-model gene regulatory network

VN#1

VN#2

VN#3

Virtual networks

VNT Agent

VNT Agent

VNT Agent

VN quality

monitoring Indirect

control

Feedback to activity

Performance

monitoring

Performance monitoring

Activity

VN: Virtual network

PN: Physical network

NTT Network Service Systems Labs.

Background

• Network Virtualization provides controllability for users

• Users freely designs own topology for adaptability

• Introduce biology-inspired topology control scheme (=self-organization)

5

PN Resource Manger

VN#1

VN#2

VN#3

VNT Agent

VNT Agent

VNT Agent

Biological system

Biological system

VN topology designed by users

NTT Network Service Systems Labs.

Adaptive VNT reconfiguration

Optical Path

IP Packets

VNT

Congestion

collect link traffic measurement

Estimate Traffic matrix

Calculate Optimized VNT

Reconfigure Optical paths

Measurement

Teardown Setup

Control PN Manager

OXC Router

VN Agent

VNT: Virtual Network Topology 1. Initial Topology

1. Optimized Topology

NTT Network Service Systems Labs.

Background: Attractor selection-based control

1. Adaptability to unknown environmental changes – System finds attractor (=equilibrium point) through random search to adapt

environmental changes.

2. Simplicity of control mechanism – Light-weight computation to find solutions

– Works well with limited information.

Gene regulatory network (=Virtual networks)

Metabolic network (=Physical network) [Kashiwagi2006] Kashiwagi et al.,”Adaptive Response of a Gene Network to Environmental Changes by Fitness-Induced Attractor Selection ,” PLoS ONE, vol.1, 2006.

Differential equation

7

NTT Network Service Systems Labs.

Outline of Talk

•Background

•Problem Statement

•Proposal: Managed self-organization

•Evaluation

•Conclusion

8

NTT Network Service Systems Labs. Our goal

•Establish resource management architecture for accommodating numerous VNs.

Some Design Principles

–Each VN is controlled according to self-organization for quick responsiveness.

•VNs can freely aquire/delete virtual resources from PN for constructing optimal topology.

–Minimal Control: Ensure enough scalability in terms of network size and the number of VNs.

9

NTT Network Service Systems Labs.

Problem we are solving

• High-performance network virtualization over optical network infrastructure.

– Wavelength path between two V-nodes fomrs virtual link in VNT.

– Dedicated wavelength resource is allocated to each VN for hard isolation.

• Key Question: Physical resource are limited. How to solve resource contention among numerous VNs? – Introduce minimal intervention for stability.

⇒ Design simple management architecture for self-organizing VNs.

10 Optical network Infrastructure L１ Switch

L2 Switch

L３ Switch

VN#3 VN#1 VN#2

Virtual Resource Allocation

contention

Self-Organization

VNT 1* VNT 2

contention

NTT Network Service Systems Labs.

Outline of Talk

•Background

•Problem Statement

•Proposal: Managed self-organization

•Evaluation

•Conclusion

11

NTT Network Service Systems Labs.

Our concept: Managed Self-Organizing System

12

Self-Organizing System (SOS)

Has an ability to operate in a dynamically

changing environment without centralized

control.

Managed Self-Organizing System

Provides a range of operating regime by an

external control while allowing a self-

organization property

Self-organizing system

Self-organizing elements

Emergence???

Self-organizing elements

Self-organizing system

Controller Observation Control

• Limitations: Applying self-organization to multiple elements (i.e., VNs) can cause instability in case of emergency.

• Our solution: Introduce minimal control in resource allocation.

Fully distributed system

Hybrid control system

NTT Network Service Systems Labs. Design approach

• How to ensure scalability of resource management? – To accommodate hundreds of VNs, we should avoid fully centralized control.

– Introduce indirect control for resolving resource contention among self-organizing VNs.

• Key observation for resource contention. – Same algorithm based on same information often cause synchronization.

⇒ Inform different residual information to each VN for diversification.

13

Under-utilized

Congested

Moderate

Moderate #2 #1

#2 #1

#2 #1

#2 #1

Initial state After control （Change topology awareness）

Link #1 Link #1

Link #1

Link #2

Link #1

Link #2

VN VN

VN #1

VN #2 VN #2

VN #1

λ paths

λ paths

NTT Network Service Systems Labs.

14

PN Resource Manager

Resource management model

Resource management system controls topology-awareness to each VN to avoid resource contention while allowing resource sharing to some resources. – Each VN utilize resources with “Right-to-use” privilege.

– By combining resources with privilege, VN topology is created on-demand.

⇒ Virtual resource layer enables indirect control of self-organizing VNs.

ＶＮ#1 ＶＮ#2

Physical Link 1 Physical Link 2

λ1

λ2

λ3

λ4 VN#1

VN#1

VN#1

VN#1

VN#2

VN#2

VN#2

VN#2 VN#2 Dedicated

VN#1,#2Shared

VN#1Dedicated

VN#2

Assign Right-to-use to each wavelength

VN#2

VN#2

VN#2 VN#2

VN#2 VN#1

VN#1

VN#1

VN#1

Link 1 2

VN is aware and can use resources with Right-to-use.

Physical resources

aware to VNs

VN

Virtual resource layer

PN VN#1

VN#1

Link 1 2

VN#2

VN#2 VN#1

VN#1

NTT Network Service Systems Labs.

Procedures for avoiding contention

15

Physical

Link 1

Physical

Link n

Lambda 1

Lambda 2

Lambda 3

Lambda 4 VN#1

VN#1

VN#1

VN#2

VN#2

VN#2

VN#2

VN#2

VN: Virtual Network

VN#1

VN#2

VN#2 VN#2

VN#1 VN#1

Shared Resource

③ Regulate resource view •Regulate residual resource view for outperformed VN. •Allocate reserved resources for congested VN.

① Initial State •All resources are shared by VNs •VN can user every resource for constructing topology

VN#1

VN#2

② Monitoring VN performance •Link load of VN is monitored. •If max load exceeds threshold, we considfer resource contention occurred.

VN#1 VN#2

NTT Network Service Systems Labs.

Outline of Talk

•Background

•Problem Statement

•Proposal: Managed self-organization

•Evaluation

•Conclusion

16

NTT Network Service Systems Labs.

Experimental Demonstration

• Developed network emulating systems for demonstrating our architecture. – Simulate behavior of multiple self-organizing VNs.

– Generate traffic demand changes and topological failures.

17

Network Emulator

PN resourceManager (DUT)

VN Agent (DUT)

Algorithm VN topologyLink load PN Resources

DUT: Device under testPN: Physical networkVN: Virtual network

Physical network

GUI System

VN Emulation

function

PN Emulator

function

Traffic Generator

VN DB

Virtual Networks

NTT Network Service Systems Labs.

Result 3: Asymptotic behavior

• Evaluate performance in simple 11-nodes Abilene topology. – SO fails to avoid resource contention.

– Proposed mechanism quickly recovered performance by reallocating resources.

18

Completely Self-organized control

allocation

Proposed (Managed SO control)

0

0.2

0.4

0.6

0.8

1

1.2

1

11

21

31

41

51

61

71

81

91

101

111

121

131

141

151

161

171

181

191

201

211

221

231

241

Traffic changes Resource reallocation

VN#1

Link load

activity

0

0.2

0.4

0.6

0.8

1

1.2

Traffic changes

VN#1

NTT Network Service Systems Labs.

Outline of Talk

•Background

•Problem Statement

•Proposal: Managed self-organization

•Evaluation

•Conclusion

19

NTT Network Service Systems Labs.

20

Concluding remarks

•Summary:

–Propose managed self-organizing network architecture

•VNs controlled based on biological systems.

•Require some mechanism for alleviating resource contention among VNs.

–Basic idea is to regulate resource view for hot spot links

–Simulation study demonstrated effectiveness of proposed algorithm

•Quickly recovered performance due to sudden traffic change

• Future work:

–Evaluate computation overhead of DRAMS.

–Evaluation in large-scale network, more than 3 VNs.

20

NTT Network Service Systems Labs.

21

Thank you

for your attention

Acknowledgement：

This work was supported in part by the SCOPE (Strategic

Information and Communications R&D Promotion Programme) of

the Ministry of Internal Affairs and Communications, Japan.