Self-stabilizing Clustering Algorithm for Ad Hoc Networks Bachar Salim HAGGAR
[email protected] SysCom, CReSTIC University of Reims Champagne-Ardenne BP1039, F-51687 Reims Cedex 2, France
August 23-29, 2009
ICWMC, 09 23-29 August 2009, Cannes
Bachar Salim HAGGAR (CReSTIC)
Clustering Algorithm
August 23-29, 2009
1 / 22
Content
1
Introduction of wireless ad hoc networks
2
Motivations for clustering
3
Clustering Principle What clustering may satisfy Previous Work
4
Our proposition
5
Conclusion and Futures works
Bachar Salim HAGGAR (CReSTIC)
Clustering Algorithm
August 23-29, 2009
2 / 22
Introduction of wireless ad hoc networks
Ad hoc Networks
Characteristics Wireless ad hoc networks consist of a set of mobile wireless nodes without fixed infrastructure Formed by wireless hosts which may be mobile and may appear/disappear at any time. Hosts can operate without any preexisting infrastructure No base station Each host acts as a router
Bachar Salim HAGGAR (CReSTIC)
Clustering Algorithm
August 23-29, 2009
3 / 22
Introduction of wireless ad hoc networks
Problematic
Problematic Movement of nodes May be unpredictable
Bachar Salim HAGGAR (CReSTIC)
Clustering Algorithm
August 23-29, 2009
4 / 22
Motivations for clustering
Motivations for clustering Why clustering MANET routing protocols are flat, thus not scalable Reduces the amount of information routing propagated in the network Can be used for routing efficiency With clustering nodes transmit their information to their clusterhead Modéling The network is modeled by a non-directed graph G(V, E) V is the set of nodes E is the set of edges {u, v } ∈ E if and only if u and v can mutually receive each others’ transmission We note Neighu the neighborhood of node u Bachar Salim HAGGAR (CReSTIC)
Clustering Algorithm
August 23-29, 2009
5 / 22
Clustering
Principle
Clustering Clustering Nodes are grouped into clusters to reduce communication overhead Clusters are identified by their clusterhead All inter-cluster communication relays through the clusterhead
Clusterhead Clusterhead
Before Clustering
Bachar Salim HAGGAR (CReSTIC)
After Clustering
Clustering Algorithm
August 23-29, 2009
6 / 22
Clustering
Principle
Clustering
What clustering may satisfy The process of cluster formation should generate as little traffic as possible The process of cluster formation should be distributed. The algorithm should be robust against nodes mobility in order to limit trafic overhead for reconstruction
Bachar Salim HAGGAR (CReSTIC)
Clustering Algorithm
August 23-29, 2009
7 / 22
Clustering
Principle
Previous Work
Previous Work Lowest-ID Cluster Algorithm [Ephremides,Wieselthier and Barker, 1987] The node with the Lowest-ID is chosen to be a clusterhead
High-Connectivity Clustering [Gerla and Lee, 1999] The node with the maximum number of neighbors is chosen to be a clusterhead
Distributed Clustering Algorithm, Distributed and Mobility-Adaptative Clustering [Besagni, 1999] The node is chosen to be a clusterhead if its node-weight is higher than any of its neighbor’s node-weights
Bachar Salim HAGGAR (CReSTIC)
Clustering Algorithm
August 23-29, 2009
8 / 22
Our proposition
Our proposition
We propose a clustering algorithm Which is completely distributed Which adapts to topology modification Based only on local information Guaranteed that no two clusterheads are neighbors With only one type of message
Bachar Salim HAGGAR (CReSTIC)
Clustering Algorithm
August 23-29, 2009
9 / 22
Our proposition
Our metric Metric Our metric is the identity Each node has an unique identity Principle The algorithm uses only one hello message to build clusters Nodes periodically exchange hello message to discover their neighborhood and make their decisions based on local information Each hello message contains only three variables: identity, cluster identity and status Hello message id
Bachar Salim HAGGAR (CReSTIC)
status
Clustering Algorithm
cl−id
August 23-29, 2009
10 / 22
Our proposition
Clusters formation
Clusters formation At the end, each node belongs to exactly one cluster Each cluster has a clusterhead, one or more gateways and zero or more ordinary nodes Each node is at a distance at most 1 of its clusterhead Gateways connect adjacent clusters
Bachar Salim HAGGAR (CReSTIC)
Clustering Algorithm
August 23-29, 2009
11 / 22
Our proposition
Clusterhead declaration Clusterhead declaration If a node u has the highest id among all its neighbors, it will become a clusterhead. if not u waits until all its neighbors with a higher id than him broadcast their decisions. If one of them is a clusterhead, u attachs itself to it and becomes an ordinary node If serveral of its neighbors are declared as clusterheads, u will become a Gateway If not, u declares itself as a clusterhead.
Status of nodes Clusterhead. Ordinaire node. Gateway. Bachar Salim HAGGAR (CReSTIC)
Clustering Algorithm
August 23-29, 2009
12 / 22
Our proposition
Clusterhead declaration
Example of clusters formation 16 4 11 2
7
13 6 5
14
9
10
Bachar Salim HAGGAR (CReSTIC)
Clustering Algorithm
1
August 23-29, 2009
13 / 22
Our proposition
Clusterhead declaration
Example of clusters formation 16 4 11 2
7
13 6 5 14
9
10
Bachar Salim HAGGAR (CReSTIC)
Clustering Algorithm
1
August 23-29, 2009
14 / 22
Our proposition
Clusterhead declaration
Example of clusters formation 16 4 11
2
7 Clusterhead
13
Gateway
6 9
Ordinary node
5
14
10
Bachar Salim HAGGAR (CReSTIC)
1
Clustering Algorithm
August 23-29, 2009
15 / 22
Our proposition
Clusterhead declaration
Example of clusters formation 16 4 11
2
7 Clusterhead
13
Gateway
6 9
Ordinary node
5
14
10
Bachar Salim HAGGAR (CReSTIC)
1
Clustering Algorithm
August 23-29, 2009
16 / 22
Our proposition
Clusterhead declaration
Example of clusters formation C16 C13 16 4 11
2
7 Clusterhead
13
Gateway
6 9
Ordinary node
5
14
1
10
C14
Bachar Salim HAGGAR (CReSTIC)
Clustering Algorithm
August 23-29, 2009
17 / 22
Our proposition
Clusterhead declaration Conflict management In case of conflict, the node that has the smallest id renounces its status of clusterhead. It reattachs to the new clusterhead. Conflict management C16 C13 16 4 11
2
7 Clusterhead
13
Gateway
6
Ordinary node 9
14
conflict
C14 10
Bachar Salim HAGGAR (CReSTIC)
Clustering Algorithm
1
August 23-29, 2009
18 / 22
Our proposition
Clusterhead declaration
Gestion de conflit C16 C9
16 4
11 2
7 Clusterhead
13
Gateway
6
Ordinary node 14
9
1
10
C14
Bachar Salim HAGGAR (CReSTIC)
Clustering Algorithm
August 23-29, 2009
19 / 22
Conclusion and Futures works
Conclusion and Futures works
Our solution A self-stabilizing algorithm. A deterministic algorithm, which builds no overlapping clusters. Convergence time is D + 2 rounds, where D is the network diameter. The size of exchanged messages is equal to Log (2n + 3). Futures works To use this algorithm in order to build a spanning tree on the clusters. To develop routing protocol based clusters. To improve clusters stability.
Bachar Salim HAGGAR (CReSTIC)
Clustering Algorithm
August 23-29, 2009
20 / 22
Conclusion and Futures works
Thak you
Thank you for your attention. Any questions??
Bachar Salim HAGGAR (CReSTIC)
Clustering Algorithm
August 23-29, 2009
21 / 22
Conclusion and Futures works
Self-stabilizing Clustering Algorithm for Ad Hoc Networks Bachar Salim HAGGAR
[email protected] SysCom, CReSTIC University of Reims Champagne-Ardenne BP1039, F-51687 Reims Cedex 2, France
August 23-29, 2009
ICWMC, 09 23-29 August 2009, Cannes
Bachar Salim HAGGAR (CReSTIC)
Clustering Algorithm
August 23-29, 2009
22 / 22
