11

Quality of resilience as a network reliability characterization tool

Cholda, P.; Tapolcai, J.; Cinkler, T.; Wajda, K.; Jajszczyk, A.;Network, IEEE

指導教授：童曉儒 老師報告人 ：林祐沁 學生日期 ： 2009/06/08

2

Outline Introduction

Relationship between QoS and QoR

Quality of Resilience

Numerical Example: The Impact of Protection Mechanisms on QoR

Conclusions

3

Introduction

Interruption could cause large loss of data the networks are equipped with relevant

mechanisms enabling survival under network failures.

A sequence of operations is necessary to perform this task: fault detection, fault localization, fault

notification, and recovery switching.

4

Introduction

Many measures have been proposed for different environments.

5

Introduction

6

Introduction

7

Introduction

Comprehensive characterization of various recovery methods implemented in multilayer networks.

Propose advanced QoS definitions, which include service resilience measures called quality of resilience (QoR).

8

Relationship between QoS and QoR

QoS is the umbrella representing the measurable (objective) requirements of the users regarding the service.bit error rate (BER), delay, packet loss

probability, available bandwidth, traffic load, and throughput.

9

Relationship between QoS and QoR

In SLAs the long-term and short-term quality characteristics can be defined.short-term quality metrics ：

related to the instant perceived quality of the service provided to the user.

long-term measure ：the overall service quality during the whole length of service operation.

10

Relationship between QoS and QoR

11

Relationship between QoS and QoR

Metrics cannot be distinguished just from quality degradation.An end user recognizes increased delay or

packet loss in the service layer leading to a certain level of TCP throughput degradation.

Those impairments might be either network congestion due to uncontrolled traffic variability, or even a hardware failure.

12

Relationship between QoS and QoR

13

Relationship between QoS and QoR

The measurement or interpretation can be performed by dividing the whole observation interval.

14

Quality of Resilience

QoR metric to compare all types of recovery strategies. It is a fractile (quantile) representation of

downtimes over the long run.

characterization purposes to report failure intensities and recovery times for management purposes.

provider-oriented approach to select recovery strategies on the basis of simulated QoR histograms.

15

Quality of Resilience

16

Quality of Resilience

QoR encompasses all of fllows:Steady-state availability and unavailability.Mean time to recovery. Mean time to failure.

T: the length of the downtime

17

Quality of Resilience The probability of uninterrupted service

The availability of the service is related to the threshold value of time α

Mean time to recovery

Mean time to failure

18

Numerical Example: The Impact of ProtectionMechanisms on QoR

Without recovery, the probability of losing the service is 0.00334.

The dedicated protection of single link/node faultsrestoration in less than 20 ms.

19

Numerical Example: The Impact of ProtectionMechanisms on QoR

Shared path protection methodsingle link faults with the usage of the

Dijkstra algorithm.

integer linear programming (ILP) to ensure the optimality of the solution

20

Numerical Example: The Impact of ProtectionMechanisms on QoR

Shared segment protectionfirst the working path is routed on the

shortest path. the second step a disjoint shared protection

route is selected with a heuristic.

21

Numerical Example: The Impact of ProtectionMechanisms on QoR

Shared protection scenarios providing resilience against dual faults.dual fault with the Dijkstra algorithm

22

Numerical Example: The Impact of ProtectionMechanisms on QoR

Mean time to recovery and some conditional fractile values are given in the description of the recovery scheme.

Fractiles can describe resilience more precisely Shared segment protection of single link faults with

recovery time constraints, MTTR = 3 min, t25%|>0 = 50 ms, t50%|>0 = 60 ms, t75%|>0 =70 ms

Dual fault with the Dijkstra algorithm, MTTR = 90 ms, t25%|>0 = 60 ms, t50%|>0 = 80 ms, t75%|>0 = 100 ms.

hreshold value like 50 ms, the standard required recovery time for SDH networks

23

Conclusions

Avoid ： choosing an improper recovery method not

adequate for user or application quality requirements.

Dealing with resilience in too narrow a scope based on only averaged metrics to describe a multidimensional quality problem.

Adopting too simple resilience agreements with customers that might lead to liability problems.