Evaluating the Trade-off between Resilience Design Alternatives in - - PowerPoint PPT Presentation

Network Architectures and Services, Georg Carle Faculty of Informatics Technische Universität München, Germany

Evaluating the Trade-off between Resilience Design Alternatives in a Virtual Network Environment under Different Network Visibility Conditions

Juan Pablo Alanis Barrera Nokia Siemens Networks Işıl Burcu Barla Stephan M. Günther 14 November, 2012

Juan Pablo Alanis Evaluating the trade-off btw resilience design alternatives in a VNE under different net. visibility conditions

2

Thesis Objective

 Design an indicator to determine at which layer the resilience provisioning should be carried out.  Evaluate the performance of the metric under different visibility conditions.  Find out the trade off between the resilience design models.

Juan Pablo Alanis Evaluating the trade-off btw resilience design alternatives in a VNE under different net. visibility conditions

3

Agenda

 Introduction to Network Virtualization

• Network Virtualization business roles
• Resilience Models

 Network Robustness indicators

• Algebraic Connectivity

 Methodology

• “Hiding Bandwidth” metric
• “QoS Differentiation” metric
• Virtual Network Simulator tool
• Example of visibility conditions

 Result Analysis and Evaluation  Conclusion

Juan Pablo Alanis Evaluating the trade-off btw resilience design alternatives in a VNE under different net. visibility conditions

4

Introduction to Network Virtualization  What is Network Virtualization?

• “Technique for isolating computational and network

resources through virtualization to allocate them to a logical (virtual) network for accommodating multiple independent virtual networks.” [1]

 Benefits

• Operating isolated network slices on diverse physical

infrastructures.

• Enabling diverse network architectures to operate on

shared physical substrates.

• Providing flexibility of adding and managing service-

tailored networks.

[1] Akiro Nakao, “Network Virtualization as Foundation for Enabling New Network Architectures and Applications”, IEICE TRANS. COMMUN., VOL.E93-B, NO.3, MARCH 2010.

Juan Pablo Alanis Evaluating the trade-off btw resilience design alternatives in a VNE under different net. visibility conditions

5

Network Virtualization Business Roles

VNO1

PIP 1 PIP 2 Mappin g Virtual node Virtual link Physical node Physical link VNet 1 VNet 2 Virtual Network Operator (VNO)

• Requests a virtual topology
• Operates the virtual topology
• Provides connectivity

Physical Infrastructure Provider (PIP)

• Possesses physical resources
• Virtualizes physical resources
• Lease resources to VNO’s

Juan Pablo Alanis Evaluating the trade-off btw resilience design alternatives in a VNE under different net. visibility conditions

6

Resilience Models

 There are two fundamental resilience models: the PIP-Resilience and the VNO-Resilience.  PIP resilience:

• The PIP is on charge of the resilience on the physical

infrastructure

• Each virtual path is mapped in 2 disjoint physical paths
• Ideally, the VNO is unaware of rerouting and failures

 VNO resilience:

• The VNO is responsible of resilience
• The working and protection path have to be physically

disjoint

• The VNO reroutes the traffic in case of failures.

Juan Pablo Alanis Evaluating the trade-off btw resilience design alternatives in a VNE under different net. visibility conditions

7

Network Robustness Indicators

 The edge connectivity η(G) of a connected graph G is the smallest number of edges whose removal disconnects G.  The vertex connectivity ν(G) of a connected graph G is the minimum number of vertices whose removal disconnects G.  The algebraic connectivity λ2(G) is defined as the second smallest eigenvalue from the Laplacian Matrix.

A D B C E = 1.38 = 2 = 2

Juan Pablo Alanis Evaluating the trade-off btw resilience design alternatives in a VNE under different net. visibility conditions

8

Algebraic Connectivity

 The algebraic connectivity λ2(G) is defined as the second smallest eigenvalue from the Laplacian Matrix L(G). Where G is graph G=(V,E).  L(G) is calculated as L = D - L.  λ2(G) is a non-decreasing function of the number of links with the same set of nodes.

2

2 D = 2 2 2

• 1
• 1
• 1
• 1

A =

• 1
• 1
• 1
• 1
• 1
• 1

Juan Pablo Alanis Evaluating the trade-off btw resilience design alternatives in a VNE under different net. visibility conditions

9

Methodology

 There is a fixed number of PIPs that provide the infrastructure to the VNO.  Each PIP has the exact same set of nodes.  The set of edges between PIPs may be different.  In each PIP physical substrate, a maximum initial load is set, that ranges from 40% to 60%.  “Hiding Bandwidth” metric

• The hiding factor h is the percentage of the link

bandwidth that is hidden by the PIP to the VNO due to resilience purposes.

• The value of h ranges from 40% to 60% of the total

link bandwidth.

Juan Pablo Alanis Evaluating the trade-off btw resilience design alternatives in a VNE under different net. visibility conditions

10

Methodology

 “Hiding Bandwidth” cont’d

• The set of visible resources to the VNO is defined as
• For this metric a threshold is set to measure the effect
• f the hidden bandwidth.
• The gain of the metric is calculated as follows:

Juan Pablo Alanis Evaluating the trade-off btw resilience design alternatives in a VNE under different net. visibility conditions

11

Methodology

 “QoS Differentiation” metric

• Three client memberships are defined C = {bronze, silver,

gold}.

• Depending on the VNO membership and the link load, the

visibility of the link is modified to the VNO:

• If a link load is less than 30%, it is shown to all clients.
• If the link load is larger than 30% and less than 70%, it is only

shown to silver and gold clients.

• If a link load is larger than 70%, it is only shown to gold clients.
• The gain of the metric is calculated as follows:

Juan Pablo Alanis Evaluating the trade-off btw resilience design alternatives in a VNE under different net. visibility conditions

12

Methodology  Virtual Network Simulator tool

• In order to obtain a benchmark of the metrics’

performance, the metric gain is compared against the

• utput of the Resilience designs of the Virtual Network

Simulator tool presented in [2].

• The VNet resilience designs are modeled as MILPs

with an objective function of delay minimization.

• The VNet delay is the sum of the observed delay of the

services running on the VNet.

• The simulator also provides the cost of the VNet, which

is defined as follows: VNet cost = (fixed + variable) for VLinks + (fixed + variable) for VNodes

[2] I. B. Barla, D. A. Schupke and G. Carle , “Resilient Virtual Network Design for End-To-End Cloud Services”, IFIP Networking 2012, Prague, May 2012.

Juan Pablo Alanis Evaluating the trade-off btw resilience design alternatives in a VNE under different net. visibility conditions

13

Example of Visibility Conditions

 Scenario: service differentiation

• Visibility conditions:
• Link load < 30%: visible to all clients
• Link load > 30% & < 70%: only visible to

silver and gold clients

• Link load > 70%: only visible to gold clients
• VNO 1 membership: bronze

VNO1

PIP 1 PIP 2 VNet 1

A B C D A B C D B A C

• PIP1 C-D link load is 50%
• The load of links from PIP2

C-D & A-C are 60% =>NO SURVIVABLE MAPPING IN EACH PIP ALONE. BASED ON THIS VISIBILITY CONDITIONS SOLUTION: Acquire resources for both PIP‘s Overlapping PIP‘s

Juan Pablo Alanis Evaluating the trade-off btw resilience design alternatives in a VNE under different net. visibility conditions

14

Result Analysis and Evaluation

 Different network sets used in the simulations

Juan Pablo Alanis Evaluating the trade-off btw resilience design alternatives in a VNE under different net. visibility conditions

15

Result Analysis and Evaluation

 This work uses the following as a proof of concept:

• 1 PIP, 1 VNO and no hidden information: the virtual

simulator output ratio is equal to 1.

• ˆ 1 PIP, 1 VNO and hidden information: the virtual

simulator output ratio is less than or equal to 1.

• ˆ n PIPs, 1 VNO and no hidden information: the virtual

simulator output ratio is higher than or equal to 1.

 Network simulator values used for the simulations.

• Total capacity on a physical links: uniform of 100 Gbps
• Bandwidth amount requested per service demand:

uniform for all services.

• Setup cost for link and nodes: proportional to the link

length.

Juan Pablo Alanis Evaluating the trade-off btw resilience design alternatives in a VNE under different net. visibility conditions

16

Results Analysis and Evaluation

 Results 7-node sparse set for Hiding Bandwidth metric  Break even point observed several times  Abrupt change of the simulator ratio seen due to the hidden information.

Juan Pablo Alanis Evaluating the trade-off btw resilience design alternatives in a VNE under different net. visibility conditions

17

Results Analysis and Evaluation

 Results 7-node mesh set for Hiding Bandwidth metric  The hidden bandwidth does not affect in the same proportion in comparison with the 7-node sparse set.

Juan Pablo Alanis Evaluating the trade-off btw resilience design alternatives in a VNE under different net. visibility conditions

18

Results Analysis and Evaluation

 Results 15-node sparse set for Hiding Bandwidth metric  Smoother change of the simulator ratio in comparison with the 7-node sparse set.  Break even point observed with high values of hidden bandwidth

Juan Pablo Alanis Evaluating the trade-off btw resilience design alternatives in a VNE under different net. visibility conditions

19

Results Analysis and Evaluation

 Results 15-node mesh set for Hiding Bandwidth metric  Break even point observed with high values of hidden bandwidth  Smaller ratios in comparison with the 15-node sparse set. As a result

• f the similar size of the edge sets in the PIP and in the VNO.

Juan Pablo Alanis Evaluating the trade-off btw resilience design alternatives in a VNE under different net. visibility conditions

20

Results Analysis and Evaluation

 Results 7-node sparse set for QoS Differentiation metric  Break even point observed only for bronze membership.  The maximum initial load values do not affect the silver and gold memberships

Juan Pablo Alanis Evaluating the trade-off btw resilience design alternatives in a VNE under different net. visibility conditions

21

Results Analysis and Evaluation

 Results 7-node mesh set for QoS Differentiation metric  Simulator output ratios equal to 1 implies that the maximum experienced delay for the services is similar in both resilience models. However, the VNet cost differs: 977 in the VNO layer and 1102 in the PIP layer.

Juan Pablo Alanis Evaluating the trade-off btw resilience design alternatives in a VNE under different net. visibility conditions

22

Results Analysis and Evaluation

 Results 15-node mesh set for QoS Differentiation metric  Maximum initial load ranges from 40% to 70%.  As the initial load increases, the experienced delays for the services at the VNO layer increases too.

Juan Pablo Alanis Evaluating the trade-off btw resilience design alternatives in a VNE under different net. visibility conditions

23

Results Analysis and Evaluation

 Service’s Bandwidth request variation for 15-node sparse set.

 As the bandwidth of the services increases, it is more complicated to the VNO-Resilience model to allocate them as the hidden bandwidth increases.

Juan Pablo Alanis Evaluating the trade-off btw resilience design alternatives in a VNE under different net. visibility conditions

24

Conclusion

 The metrics‘ performace shows correctly at which layer the resilience provisioning should be carried

• ut, based on the optimization objective.

 The metric provides a quick and efficient insight at which layer the resilience should be realized.  For the simulator ratios equal to 1, it is necessary to look into another network parameter to determine at which layer is more convinient.  For the same set of networks, it can be said that the metric gain with a higher value implies a lower experienced delay for the services at the VNO layer.

Juan Pablo Alanis Evaluating the trade-off btw resilience design alternatives in a VNE under different net. visibility conditions

25

Questions

Thank You for Your Attention Questions?