On Exploiting Diversity for Cluster Formation in Self-Healing MANETs - - PowerPoint PPT Presentation

On Exploiting Diversity for Cluster Formation in Self-Healing MANETs

Ann T. Tai Kam S. Tso IA Tech, Inc., USA William H. Sanders Information Trust Institute University of Illinois, USA September 17, 2009

PMCCS-9 (Eger, Hungary) Cluster Formation in Self-Healing MANETs September 17, 2009 1 / 13

Introduction

Self-organizing ability is the key to self-healing networks, especially MANETs (mobile ad hoc wireless networks). The better the performance and robustness of a self-organizing mechanism, the better survivability of a network system. Clustering approach, a self-organizing mechanism, enables not only MANET scalability but also survivability.

PMCCS-9 (Eger, Hungary) Cluster Formation in Self-Healing MANETs September 17, 2009 2 / 13

Basics of Clustering

PMCCS-9 (Eger, Hungary) Cluster Formation in Self-Healing MANETs September 17, 2009 3 / 13

Traditional Clustering Approaches: Merits

Majority of them are decentralized algorithms. Cluster formation results in a communication hierarchy, enabling network scalability.

PMCCS-9 (Eger, Hungary) Cluster Formation in Self-Healing MANETs September 17, 2009 4 / 13

Traditional Clustering Approaches: Inadequacies

Performance is heavily dependent upon network topology. Iterative execution is necessary for achieving satisfactory clustering coverage (the fraction of nodes that become organized through clustering). Cluster maintenance, which is often complicated and costly, are required to ensure persistent coverage. Hybrid/adaptive clustering methods apply alternate clustering policies conditionally (upon the detection of a poor coverage or CH-election conflict), limiting their performance gain.

PMCCS-9 (Eger, Hungary) Cluster Formation in Self-Healing MANETs September 17, 2009 5 / 13

Our Approach

Simultaneous application of diversified clustering policies. Complementary, superimposed cluster-layers resulting from diversified policies. A single-round protocol. Affordable partial redundancy pruning (no additional message exchanges required).

PMCCS-9 (Eger, Hungary) Cluster Formation in Self-Healing MANETs September 17, 2009 6 / 13

Superimposed Clustering: A MaxMin-Based Instance

Scenario 1:

16 20 13 24 26 21 12 15 1 18 5 25 4 9 10 19 8 7 6 2 29 23 11 22 28 14 27 17 3

(a) By Max-ID Policy

16 20 13 24 26 21 12 15 1 18 5 25 4 9 10 19 8 7 6 2 29 23 11 3 22 28 14 27 17

(b) By Min-ID Policy

16 20 13 24 26 21 12 15 1 18 5 25 4 9 10 19 8 7 6 2 29 23 11 22 28 14 27 17 3

(c) By MaxMin-ID Policy

Scenario 2:

16 13 20 2 15 22 24 6 25 9 29 26 18 1 27 14 4 19 7 21 8 3 28 5 12 23 11 10 17

(a) By Max-ID Policy

16 13 20 2 15 22 24 6 25 9 29 26 18 1 27 14 4 19 7 21 8 3 28 5 12 23 11 10 17

(b) By Min-ID Policy

16 13 20 2 15 22 24 6 25 9 29 26 18 1 27 14 4 19 7 21 8 3 28 5 12 23 11 10 17

(c) By MaxMin-ID Policy

PMCCS-9 (Eger, Hungary) Cluster Formation in Self-Healing MANETs September 17, 2009 7 / 13

SCP Algorithm

1: send(nID(v));

// ID diffusion

2: N1(v) ← receive(integerSet);

// N1 is a set of nIDs of 1-hop neighbors of v

3: send(nID(v),N1(v));

// neighborhood info exchange

4: N2(v) ← receive(2-tupleSet);

// N2 is a set of n ID, N1

5: CHs(v) ← identifyCH(N2(v)); 6: CHstatus(v) ← redundancyChecking(nID(v), MAXorMIN(nID(v),N1(v)), N1(v),N2(v)); 7: send(nID(v), CHs(v), CHstatus(v), nil); 8: cRegistry(v) ← receive(4-tupleSet); 9: CHstatus(v) ← redundancyPruning(CHstatus(v), cRegistry(v)); 10: identifyGW(cRegistry(v));

PMCCS-9 (Eger, Hungary) Cluster Formation in Self-Healing MANETs September 17, 2009 8 / 13

Principles of SCP

Minimal message exchanges among the nodes: MaxMin-ID-based algorithm involves three message exchanges, the same as the Max-ID (or Min-ID) based algorithms. Emphasis on local computation: Neighborhood analysis, role identification, and redundancy checking/pruning are all performed locally at individual nodes.

As message exchange is almost always the major drive of performance

• verhead and energy consumption in MANETs, the SCP approach can

be justified by both its efficiency and affordability.

PMCCS-9 (Eger, Hungary) Cluster Formation in Self-Healing MANETs September 17, 2009 9 / 13

An Analytic Model

P(C2) = PImax

i=1 P(ID(v) = i) PImax −1 n=1

P(| N1(v) |= n) P(Gh(v) = 1 ∨ Gm(v) = 1 | ID(v) = i, | N1(v) |= n) P(C2) =

Imax

X

i=1

1 Imax

Imax −1

X

n=1

“Imax − 1 n ” „ Av At «n „ 1 − Av At «(Imax −1)−n `i−1

n

´ `Imax −1

n

´ + `Imax −i

n

´ `Imax −1

n

´+

min {n−1,Imax −i}

X

m=max{1,n−(i−1)}

`Imax −i

m

´` i−1

n−m

´ `Imax −1

n

´ @1 − @

Imax

X

j=i+1

1 Imax − i 1 − „ 1 − Au At «Imax −j!1 A

m

i−1 X

k=1

1 i − 1 1 − „ 1 − Au At «k−1!!n−m1 A

PMCCS-9 (Eger, Hungary) Cluster Formation in Self-Healing MANETs September 17, 2009 10 / 13

Evaluation Results: Clustering Coverage as a Function of r

0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 10 15 20 25 30 35 40 45 50 Clustering Coverage Transmission Range Max only MaxMin

(a) Low Node-Population Density

0.65 0.70 0.75 0.80 0.85 0.90 0.95 10 15 20 25 30 35 40 45 50 Clustering Coverage Transmission Range Max only MaxMin

(b) High Node-Population Density

PMCCS-9 (Eger, Hungary) Cluster Formation in Self-Healing MANETs September 17, 2009 11 / 13

Summary

We exploit parallelism and diversity in a novel fashion:

◮ We let two different layers of clusters be formed in parallel to compose

a significantly better coverage.

◮ SCP applies complementary clustering policies to achieve a better

clustering coverage by taking advantage of result diversity.

The SCP framework provides mobile wireless hosts with a low-cost self-organizing capability which is the key to self-healing MANETs.

PMCCS-9 (Eger, Hungary) Cluster Formation in Self-Healing MANETs September 17, 2009 12 / 13

Future Work

Formalize the SCP framework. Perform redundancy analysis: To view and utilize it in a positive way. Evaluate measures taking into account network dynamics. Allow nodes to move, fail, or die amid clustering.

PMCCS-9 (Eger, Hungary) Cluster Formation in Self-Healing MANETs September 17, 2009 13 / 13