Elsevier

Computer Networks

Volume 52, Issue 9, 26 June 2008, Pages 1797-1824
Computer Networks

Clustering in mobile ad hoc networks through neighborhood stability-based mobility prediction

https://doi.org/10.1016/j.comnet.2008.01.018Get rights and content

Abstract

Clustering for mobile ad hoc networks (MANETs) offers a kind of hierarchical organization by partitioning mobile hosts into disjoint groups of hosts (clusters). However, the problem of changing topology is recurring and the main challenge in this technique is to build stable clusters despite the host mobility. In this paper, we present a novel clustering algorithm, which guarantees longer lifetime of the clustering structure in comparison to other techniques proposed in the literature. The basis of our algorithm is a scheme that accurately predicts the mobility of each mobile host based on the stability of its neighborhood (i.e., how different is its neighborhood over time). This information is then used for creating each cluster from hosts that will remain neighbors for sufficiently long time, ensuring the formation of clusters that are highly resistant to host mobility. For estimating the future host mobility, we use provably good information theoretic techniques, which allow on-line learning of a reliable probabilistic model for the existing host mobility.

Introduction

Clustering [1] is a promising approach for enhancing the scalability of mobile ad hoc networks (MANETs) in the face of frequent topology changes mainly due to the host mobility. Clustering not only makes a large MANET to appear smaller, but more importantly, it makes a highly dynamic topology to appear less dynamic [3]. In clustering, a representative of each cluster is ‘elected’ as a cluster head (CH) and a mobile host (MH), which serves as intermediate for inter-cluster communication, is called a gateway. Remaining members are called ordinary MHs. CHs hold routing and topology information while the boundaries of a cluster are defined by the transmission area of its CH.

The feasibility of a clustering method is determined by the stability of the cluster structure that it creates, despite network topology changes. Otherwise, frequent reclustering is required thereby creating a large volume of control messages which in turn consume considerable bandwidth and drain MHs’ energy quickly. As the main cause for topology changes in MANET is the host mobility, an efficient clustering method should seriously take the movements of MHs into account in order to form clustering structures resistant to the host mobility.

Many researchers [3], [4], [5], [6], [9], [10], [11], [12] have acknowledged the importance of host mobility estimation for building clustering schemes more stable and less reactive to topological changes of ad hoc networks. Authors in [3] propose the (a,t) clustering scheme, where MHs form clusters according to a path availability criterion. The network is partitioned into clusters of MHs, that are mutually reachable along cluster internal paths which are expected to be available for a period of time t with a probability of at least a. The parameters of this model are predefined. In addition, it is assumed that the movement of each MH is random and entirely independent of the movements of other MHs. However, this random walk model cannot always capture some host mobility patterns occurring in practice in MANETs.

MOBIC in [4] elects as CHs the MHs which exhibit the lowest mobility in their neighborhood. Each MH compares the receiving signal strength from its neighbors over the time and uses the variance in these values as an indication of how fast this MH is moving in relation to the neighboring MHs. MOBIC uses only the current mobility to determine the most suitable MHs for CHs. As an extension of MOBIC, MobDHop [5], [6] also uses the variability in receiving signal strength as a hint of neighborhood mobility and builds variable-diameter clusters. It uses more samples of receiving signal than MOBIC to estimate the predicted mobility but again the prediction model is rather simple since it is based on the assumption that the future mobility patterns of MHs will be exactly the same as those of the recent past.

MAPLE [7] is another clustering algorithm which also infers host mobility from measurements of the received signal strength. In particular, each MH belonging to a cluster can estimate its distance from its CH by using the well-known formula [8] of the signal attenuation versus travelled distance. Then, based on past measurements, MHs use a linear model for estimating their future distance from their CH. This helps MHs proactively join another cluster if they are going to leave their current cluster. However, MAPLE does not take host mobility into account during CH election. Specifically, MHs contend for free frames (i.e. time slots) in a single shared broadcast channel during the cluster formation phase and MHs that first reserve the available frames in this phase become CHs. Thus, the election of CHs is mostly a random procedure and it is not based on some CH suitability criteria. In addition, the algorithm sets an upper bound on the number of clusters in the network as well as on the number of MHs per cluster.

DMAC in [9] and GDMAC in [10] proposed by Basagni are generic weight-based clustering schemes, where MHs with the highest weight among neighboring ones are elected as CHs. Basagni suggested to use the inverse of the speed of MHs as a weight in its scheme. WCA in [11] is also a weight-based clustering technique which extends the work in [9], [10]. The weight in this scheme is determined by considering various factors that affect the suitability of a MH as a CH. Among these factors is host mobility. Specifically, each MH measures its average speed by sampling its position coordinates at regular time intervals. This method of measurement requires the use of a GPS device on each MH, which is not always feasible. Furthermore, this method fails to capture the correlation that may exist among the movements of neighboring MHs as in the case of group movement.

Information theory-based techniques for host mobility prediction have been first employed in [14], where the authors focused on the problem of mobile tracking and localization in cellular networks. Later, Sivavakeesar et al. [12] used the basic technique of [14] in their cluster formation algorithm for MANETs. A basic assumption in their work is that a geographical area is divided into circular-shaped regions named virtual clusters and each MH knows the virtual cluster where it is currently in. So, the ad hoc network in their technique is very much like a cellular one and the ideas in [14] can be applied.

In this paper, we propose a novel mobility-aware technique for cluster formation and maintenance. The main idea in our technique is to estimate the future mobility of MHs so as to select CHs that will exhibit the lowest predicted mobility in comparison to the other MHs. As a measure of host mobility rate, we use the probability of a MH having the same MHs in its neighborhood for sufficiently long time. A high probability value for a MH indicates a relatively immobile host or the existence of a group of MHs around this particular MH that exhibits the same mobility pattern. Whatever the case is, this MH is apparently a good candidate for a CH, because in all probability, it will serve the same neighbors for a long time. For estimating the predicted mobility of a MH, we make the realistic assumption for most MANETs that the movements of MHs are not random but demonstrate a regular pattern [15], which can be predicted provided that enough “historic” information has been gathered for the movements of each MH.

For the organization of the historic record and the estimation of future mobility based on this record, we borrow prediction techniques from the field of data compression. Specifically, we reduce the problem of estimating the future neighborhoods of a MH to one of predicting the next characters in a text given that we have already seen a particular text context. Then, by using context-modelling techniques [16], we can reliably estimate the probability of stable neighborhood around a MH. The most important characteristic of these methods is the on-line learning of the probability model which these methods use for prediction of the next character/neighborhood. This is essential in the case of ad hoc networks because the movements of individual MHs as well as the strong correlation that exists in the movements of these hosts cannot be easily described by predefined random models.

Note also that we do not make any use of a fixed geographical partition in contrast to previous work [14], [12] and thus the notion of cells is irrelevant to our technique.

Besides the stability of the clustering structure, an important objective in cluster creation is to keep the number of elected CHs relatively low so that the virtual backbone built over these MHs will be of correspondingly small-size and hence routing update protocols could be efficiently ran on this backbone. The well-known highest connectivity (degree) algorithm [2] promises the election of relatively few CHs. In this paper, we propose a new clustering algorithm named MobHiD, which combines the highest degree technique with our mobility prediction scheme and ensures a relatively small as well as stable virtual backbone despite host mobility. The performance of our technique was verified via simulation experiments, which compared our algorithm with other competitive techniques of the literature.

Note that our mobility prediction technique is of independent interest and may be combined with other clustering algorithms to enhance the stability of the derived clustering structure in the presence of frequent topology changes.

The paper is organized as follows. In Section 2, we discuss our mobility prediction method. In Section 3, we present our MobHiD clustering algorithm which uses the mobility prediction method. Section 4 addresses the details of the distributed implementation of MobHiD and then Section 5 discusses the simulation results about the performance of our clustering technique. Finally, Section 6 concludes the paper by summarizing the main contribution of our work.

Preliminary results of this work have been presented in [20].

Section snippets

Our mobility prediction method

A MH is considered a good candidate for CH if its neighborhood is relatively stable in comparison to the neighborhoods of other candidate hosts. Let nghi,t={h0,h1,h2,,hnt-1} be the nt neighboring MHs of MH i at time step t. Somehow, we have to estimate the probability of the stability of this neighborhood, i.e., the probability P(nghi,t) that this neighborhood will remain the same in the following time steps, if possible, forever. By making the simplified assumption that the presence of a MH

Mobility-aware highest degree (MobHiD) technique

The distributed implementation of our technique should create a stable clustering structure with minimal control overhead. As CHs and gateways will form the virtual backbone through which messages will be routed on the ad hoc network, the size of this backbone should be kept as small as possible so that the delay of message routing is correspondingly small. Also, we opt for one-hop clusters where each MH is one-hop away from its CH. In this way, the routing decisions inside each cluster are

Distributed implementation

In the proposed mobility prediction method, each MH i should compute its weight wi according to the weight formula (5). Therefore, each MH should know its neighbors and how its neighborhood changes over time. This implies a periodic exchange of HELLO messages, namely messages HELLO(clusterhead?,wi) so that each MH i can inform its neighbors about its presence, whether it is a clusterhead or not and about its weight. The information carried by the two fields of the HELLO message proves useful

Simulation results

The performance of the MobHiD algorithm was tested through a series of simulation experiments on the ns2 simulator [26]. For comparison, we also simulated four other one-hop clustering algorithms, namely the Lowest ID (LI) [2], Highest Degree (HD), GDMAC as well as MOBIC. First, we measured how the number of created clusters/CHs of each technique varies with the total number of MHs. Then, we studied the variation of lifetime (duration) of elected CHs in each of these methods against the maximum

Conclusions

In this paper, we presented a mobility-aware clustering scheme which uses well known information theoretic techniques for reliably estimating the future mobility of MHs. The right prediction enables the formation of clusters that are highly resistant to the topological changes of the ad hoc network due to host mobility. For measuring the mobility, we do not use special purpose hardware such as GPS but the mobility of each host is inferred from how different is the neighborhood of the MH over

Charalampos Konstantopoulos received his Diploma in computer engineering from the Department of Computer Engineering and Informatics at University of Patras, Greece (1993). He also received his Ph.D., degree in Computer Science from the same department in 2000. Currently, he is a Ph.D., researcher at the Research Academic Computer Technology Institute, Patras, Greece. His research interests include parallel and distributed algorithms/architectures, mobile computing and multimedia systems.

References (29)

  • I. Er et al.

    Performance analysis of mobility-based d-hop (MobDHop) clustering algorithm for mobile ad hoc networks

    Computer Networks

    (2006)
  • E. Dantsin et al.

    A deterministic (2-2/(k+1))n algorithm for k-SAT based on local search

    Theoretical Computer Science

    (2002)
  • J. Yu et al.

    A survey of clustering schemes for mobile ad hoc networks

    IEEE Communications Survey

    (2005)
  • M. Gerla et al.

    Multicluster, mobile, multimedia radio network

    ACM-Baltzer Journal of Wireless Network

    (1995)
  • A. McDonald et al.

    A mobility-based framework for adaptive clustering in wireless ad hoc networks

    IEEE Journal on Selected Areas in Communications

    (1999)
  • P. Basu, N. Khan, T. Little, A mobility based metric for clustering in mobile ad hoc networks, in: Proceedings of the...
  • I. Er, W. Seah, Mobility-based d-hop clustering algorithm for mobile ad hoc networks, in: Proceedings of IEEE Wireless...
  • R. Palit et al.

    MAPLE: a framework for mobility-aware pro-active low energy clustering in ad-hoc mobile wireless networks

    Wireless Communications and Mobile Computing

    (2006)
  • M. Schwartz

    Mobile Wireless Communications

    (2005)
  • S. Basagni, Distributed clustering for ad-hoc networks, in: Proceedings of the 1999 International Symposium on Parallel...
  • S. Basagni, Distributed and mobility-adaptive clustering for multimedia support in multi-hop wireless networks, in:...
  • M. Chatterjee et al.

    WCA: a weighted clustering algorithm for mobile ad hoc networks

    Cluster Computing

    (2002)
  • S. Sivavakeesar, G. Pavlou, A. Liotta, Stable clustering through mobility prediction for large-scale multihop ad hoc...
  • D. Johnson et al.

    Dynamic source routing in ad hoc wireless networks

  • Cited by (59)

    View all citing articles on Scopus

    Charalampos Konstantopoulos received his Diploma in computer engineering from the Department of Computer Engineering and Informatics at University of Patras, Greece (1993). He also received his Ph.D., degree in Computer Science from the same department in 2000. Currently, he is a Ph.D., researcher at the Research Academic Computer Technology Institute, Patras, Greece. His research interests include parallel and distributed algorithms/architectures, mobile computing and multimedia systems.

    Damianos Gavalas received his BSc degree in Informatics from University of Athens, Greece, in 1995 and his MSc and Ph.D., degree in electronic engineering from University of Essex, UK, in 1997 and 2001, respectively. He is currently Assistant Professor in the Department of Cultural Technology and Communication, University of the Aegean, Greece. His research interests include distributed computing, mobile code, network and systems management, network design, e-commerce, m-commerce, mobile ad hoc and wireless sensor networks.

    Grammati Pantziou received the Diploma in Mathematics and her Ph.D. degree in Computer Science from the University of Patras, Greece, in 1984 and 1991, respectively. She was a Post-Doctoral Research and Teaching Fellow at the University of Patras (1991–1992), a Research Assistant Professor at Dartmouth College, Hanover, NH, USA (1992–1994), an Assistant Professor at the University of Central Florida, Orlando, FL, USA (1994–1995) and a Senior Researcher at the Computer Technology Institute, Patras (1995–1998). Since September 1998, she is a Professor at the Department of Informatics of the Technological Educational Institution of Athens, Greece. Her current research interests are in the areas of parallel computing, design and analysis of algorithms, distributed and mobile computing and multimedia systems.

    This work is co-funded by 75% from EU and 25% from the Greek Government under the framework of the Education and Initial Vocational Training II, programme “Archimedes”.

    View full text