Survey of clustering algorithms for MANET

Ratish Agarwal et al / Internat ional Jo urnal on Com puter Science and Engineering Vol.1(2), 2009, 98-104 98 Survey of clustering algorithms for MANET Ratish Agarwal, 1 Departm ent of Inform ation T echnology, Rajiv Gandhi Pro udyogiki V ishwavidyal aya, Bhopal ratish@rgtu.ne t Dr. Mahesh Motwani Departme nt of Computer Sci ence Enginee ring, Jabalpur Engi neering Colle ge , Jabalpur mahesh_9@h otmail.com Abstract - Many clustering scheme s have been proposed for ad hoc networ ks. A systematic c la ssification of these clustering schemes enables one to better understand and make improvements. In mobile ad hoc networks, the movement of the network nodes may quickly change the topology resulting in the increase of the overhead message in topology maintenance. Protocols try to keep the number of nodes in a cluster around a pre-defined threshold to facilitate the optimal operation of the medium access con trol protocol. The clu sterhead el ection is invoked on-demand, and is aime d to reduce the computation and communication costs. A large var iety of approaches for ad hoc clustering have be en developed by researc hers which focus on different performan ce metrics. This pap er presents a surve y of different clustering schemes . Keywords: Mobile ad hoc netw orks, Clustering, clusterhead, gateway. I. I NTRODUCTION In an ad hoc network, mob ile nodes communicate with each other using mult ihop wireless li nks. There is no stationary infrastructure; for instance, there are no base stations. Each node in the network also acts as a router, forwardi ng data packets for other nodes. A re sear ch issue in the design of ad hoc networks is t he development of dynam ic routing protocols that can efficiently find rout es between two communicating nodes. The routi ng protocol m ust be able to keep up wit h the high degree of no de mobility that often changes the net work topology. In a l arge network, flat rout ing schemes produce an excessive amount of info rmation that can saturat e the network. In addition, given t he nodes heterogeneity, nodes m ay have highly variabl e amount of resou rces, and this produce s a hierarchy in their ro les inside the netw ork. Nodes w ith large computati onal and comm unication power, and powerful batteries are more suitabl e for supporting th e ad hoc network functions (e.g., routing) than o ther nodes. Cluster-based routing is a solution to address nodes heterogeneity, and t o limit the am ount of routing i nformati on that propagates insi de the network. The idea behind clust ering is to group th e network nodes i nto a number of overl apping clusters. Clustering m akes possible a hierarchical routing i n which paths are recorded between clusters instead of between nodes. This increases the routes lifetime, thus decreasing the amount of rout ing control overhead. Inside th e cluster one node that coordinates the clu ster activities is clusterhead (CH). Inside the cluster, there are ordinary nodes also t hat have direct access only to this one clusterhead, and gateways. Gateways are nodes that can hear two or more clusterheads. Ordinary node s send the packets to their cl usterhead that either distributes the packets in side the cluster, or (if the destination is o utside the cluster) forward s them to a gateway node to be delivered to th e other clusters. By replac ing the nodes with cluste rs, existing routi ng protocol s can be directly applied to the n etwork. Only gateways an d clusterheads participate in the propag ation of routin g contro l/update messages. In dense networks this significantly reduces the routing overhead, thu s solving scalability prob lems for routing algorithms in large ad hoc networks. II. C LUSTERING A LGORITHMS IN MANET We present below a sur vey of different cl ustering algorithms. 2.1 Identifier-based clustering A unique ID is assigned to each node. No des know the ID of its neighbors and cl usterhead is chosen fol lowing some certain rules as given below. 2.1.1 Lowest ID cluster algorithm (LIC) [1] is an algorithm in wh ich a node with the minimum id is chosen as a clusterhead. Thus, the ids of the neighbors of t he clusterhead will be higher than that of the clusterhead . A node is called a gateway if it lies with in the transmission rang e of two or more clusterheads. Gateway nodes are generally used for routing between cluste rs. Each node is assigned a distinct id . Periodically, the n ode broadcasts the list of n odes that it can hear (including i tself ). • A node which only hears nodes wit h id higher than itself is a clusterhead. • The lowest- id node that a node hears is its clusterhead, unless the lowest- id specifically gives up its role as a clusterhead (deferring to a yet lower id node). • A node which can hear two or m ore clusterheads is a gateway. • Otherwise, a node is an ordi nary node. The Lowest-ID schem e concerns only with the lowest no de ids which are arbitrarily assigned n umbers without consideri ng any other qualification s of a no de for election as a clusterhead. Since the node ids do not chan ge with time, those with s maller ids are more likely to become clusterheads than nodes with larger ids . Thus, drawback of lowest ID algori thm is that certain nodes are prone to power drainage due to serving as clusterheads for longer periods of tim e. ISSN : 0975-3397 Ratish Agarwal et al / Internat ional Jo urnal on Com puter Science and Engineering Vol.1(2), 2009, 98-104 99 2.1.2 Max-Mi n d-cluster formati on algorithm [2] generalizes the cluster defini tion to a coll ection of nodes that are up to d-hops away from a clusterhead. Due to the large number of no des involve d, it is desirable to let the no des operate asynchron ously. The clock sync hronization o verhead is avoided, providing ad ditional processing saving s. Furthermore, t he number of messages sent from each node is limited to a multip le of d the m aximum num ber of hops away from the nearest clusterhead, rather tha n n the number of nodes in the n etwork. This guaran tees a good controlled message complexity for the al gorithm . Additionally, because d is an input value to the heuristic, there is control over the number of clusterheads elected or t he density of cl usterheads in the network . The amount of resources n eeded at each nod e is minimal, con sisting of four simple rules and tw o data structures that maintain node information over 2d rounds of communicat ion. Nodes a re candidate s to be clust erheads b ased on their no de id rather than their degree of connectivity. As the network topol ogy changes slightly t he node’s degree of connectivity is much more lik ely to change than th e node’s id relative to its neigh boring nodes. If a node A is the largest in the d-neighb orhood of another node B then node A, A will be elected a clusterhea d, even though node A may not be the largest in it s d-neighbor hood. This p rovides a smoot h exchange of clusterheads rather than an erratic excha nge. This method m inimizes the am ount of dat a that m ust be pas sed from an outgoing cl usterhead to a new clusterhead when there is an exchange. 2.2 Connectivity-based clustering 2.2.1 Highest connectivity c lustering algorithm (HCC) [1] The degree of a no de is computed based on it s distance from others. Each node broa dcasts its id to the nodes that are within its tran smission rang e. The node with maxi mum number of neighbors (i.e., maxim um degree) is chosen as a clusterhead. The neighbors of a clusterhead b ecom e members of that cluster and can no longer participate in the election process. Since no clusterheads are directly linke d, only one clusterhead is allowed per cluster. Any t wo nodes in a cluster are at most two- hops away since the clusterhead is directly linked to each of its neighbors in the cluster. Basica lly, each node either becomes a clusterhead or rem ains an ordinary node. This system has a low rate of clusterhead chang e but the throughput is low. Typically, each cluster is assigned some resources which is shared among the members of that cluster. As the number of nodes in a clus ter is increased, the throughput drops. The reaffiliatio n count of nod es is high due to node movement s and as a result, the highest -degree node (t he current clusterhead) may not be re-elected to be a clusterhead even if it looses one neighbor. All these drawbacks occur because this approach does not have any restri ction on the upper bound on the number of nodes i n a cluster. 2.2.2 K-hop connectivity I D clustering algorithm (K- CONID) [3] combines two clusterin g algorithms: the Low est- ID and the Highest -degree heurist ics. In order to se lect clusterheads connectivity is cons idered as a first criterion and lower ID as a secondary cri terion. Using only node connectivity as a criterion causes numerous ties between nodes On the other hand, usi ng only a lower ID cri terion generat es more clusters than necessary. The purpose i s to minimize the number of clus ters formed in the net work and in this way obtain dominati ng sets of smaller si zes. Clusters in the K- CONID approach are formed by a clusterhead and all nodes that are at distance at most k-hops from the clusterhead. At the beginning of the algorithm , a node starts a flooding process in which a clust ering request is send to all other nodes. In the Highest-degree heuristic , node degree only measures connectivity for 1- hop clust ers. K-CONID ge neralizes connectivity for a k-hop nei ghborhood. Thus, whe n k = 1 connectivity is the same as node degree. Each node in the network is assigned a pair di d = (d, ID). d is a node’s connectivit y and ID is the nod e’s identifier. A node is selected as a clusterh ead if it has the highest co nnectivity. In case of equal connectivity, a node has clusterhead priority if it has lowest ID. The basic idea is that every node broadcasts its clustering decision once al l its k-hop neighbors wit h larger clusterhead priorit y have done so. 2.2.3 Adaptive cl uster load balance method [4]. In HC C clustering scheme, one cl uster head can be exhausted when it serves too many mobile hosts. It is not desirab le and the CH becomes a bottleneck. So a new appr oach [4] is given. In he llo message format, there is an "Option" ite m. If a sender node is a cluster head, it w ill set the nu mber of its domin ated member nodes as "Option" val ue. When a sender node is not a cluster head or it is undecided (CH or non -CH), "Option" item will be reset to 0. When a CH 's Hello message sh ows its dominated nodes' number exceeds a thresh old (the maxim um number one CH can manage), no n ew node will partici pate in this clu ster. As a result, this can eli minate the CH bo ttleneck phenomenon and optimize the cluster structure. This algorithm can get load balance between various clust ers. Thus, resource consumpti on and information tran smission is distributed to all clu sters instead of few clusters. 2.2.4 Adaptive multihop clustering [5] is a multihop clustering scheme with load-balancing capabilit ies. Each mobile node p eriodically broadcasts informati on about its ID, Clusterhead ID, and its status (c lusterhead/mem ber/gateway) to others within the same cluster. W ith the help o f this broadcast, each mobile node obtains the topology information of its cluster. Each gateway also peri odically exchanges information with neighbor ing gateways in d ifferent clusters and reports to its clusterhead. Thus, a cluste rhead can know the num ber of mobile nodes of each neighboring cluster. Adaptive m ultihop clustering sets uppe r and lower bounds (U and L ) on the number of clustermembers within a cluster that a clusterhead can handle. When the number of clustermembers in a cluster is less than the lowe r bound, the cluster n eeds to merge with o ne of the neighboring clust ers. In order to merge two clust ers into one cluster, a clusterhead always has to get the cluster size of all neighboring clusters. It prevents th at the number of clustermembers in the merged clu ster is over the upper bound. On the contrary, if the number of clustermembers in a cluster is greater than the up per bound, the cluster is d ivided into two clusters. However, Ada ptive multi hop clustering does not address how to select a proper no de to serve as the clusterhead ISSN : 0975-3397 Ratish Agarwal et al / Internat ional Jo urnal on Com puter Science and Engineering Vol.1(2), 2009, 98-104 100 for the newly detached cluster. The upper and lower bounds are decided by network si ze, mobility etc. 2.3 Mobility-aware clustering 2.3.1 Mobility-based d-hop clustering algorithm [6] partitions an ad h oc network into d-h op clusters based on mobility metric. The objective of form ing d-hop cluste rs is to make the clus ter diameter more flexible. This algorith m is based on mobilit y metric and the diam eter of a cluster i s adaptable with respect to node m obility. This clustering algorithm assumes that each no de can measure its received signal strength. In this manner, a node can estimate its distance from its neighbors. Strong r eceived signal strength im plies closeness between two nodes. T his algorithm requires the calculation of f ive terms: the est imated distance between no des, the relative mobilit y between node s, the variation of esti mated distance over time, the local stab ility, and the estimated mean distance. Relative mobility corresp onds to t he difference of t he estimated distance of one node w ith respect to another, at two successive time moments. T his parameter indicates if two nodes move away from each other or if they become closer. The variation of estim ated distances between two n odes is computed instead of cal culating physical distance between two nodes. This is because physical distance between two nodes is not a precise measure of closeness. For instance, if a node r uns out of energy i t will t ransmit packets at low power acting as a distanced node from its physicall y close neighbor. The variation of estima ted distance and the relativ e mobility between nodes are used to calcul ate the local st ability. Local stability is computed in order to select some nodes as clusterheads. A node m ay become a clusterhead if it is found t o be the most stable no de among its neighb orhood. Thus, the clusterhead will be the node with the lowest value of local stability am ong its neighbo rs. 2.3.2 Mobility Based Metric for C lustering [7] proposes a local mobility metr ic for the cluster formation pro cess such that mobile nodes wi th low speed relati ve to their neighbors have the chance to become clusterheads. By calculating the variance of a mobile node’s speed relativ e to each of its neighbors, the aggregate local speed of a m obile node is estimated. A low variance value indicates that this mobile node is relatively less mobile to its nei ghbors. Consequently , mobile nodes with low variance values in their neighbor hoods are chosen as clusterhead. For cluster maintena nce, timer is used to re duce the clusterhead change rate by avoiding re-clust ering for incidental contacts of two passing cl usterheads. However, the mobility behavi or of mobile nodes is not always consi dered in cluster maintenance, so a cluste rhead is not guaranteed to bear a low mobility characteristic relative to its members durin g maintenance phase. As time advances, the m obility criterion is somewhat ignored. T his scheme is effective for M ANETs with group mobil ity behavior, in whic h a group of m obile nodes moves with simil ar speed and direction, as in highway traffic. Thus, a selected clusterhead can normally promise the low mobility wi th respect to its member nodes. However, if mobil e nodes move random ly the performance m ay reduce. 2.3.3 Mobility-based Frame Work fo r Adaptive Clustering [8] p artition a number of mobile n odes into multi- hop clusters based on (a, t) cri teria. The (a, t) criteria indicate that every mobile node in a cl uster has a path to every other node that will be availa ble over some time peri od ‘t’ with a probability ‘a’ regard less of the hop distance betw een them. Cluster framework is based on an a daptive architecture designed to dynam ically organize m obile nodes into cluster s in which the probabi lity of path avail ability can be bounded, a nd the impact of routing overhead can be effectively managed. The cluster organizati on supports an adapt ive hybrid rout ing strategy that is more respon sive and effective when node mobility is low an d more efficient when node mobilit y is high. The purpose of this strategy is to su pport a more scalable routing infrastructur e that is able to adapt to h igh rates of topological change. T his is achie ved using predict ion of the future state of the network li nks in order to provide a quantitative b ound on th e availability of paths to clu ster destinations. A metric which captu res the dynamics o f node mobility, mak es the scheme ad aptive with respect to node mobility. 2.4 Low cost o f main tenance clustering 2.4.1 Least cluster change algorithm (L CC) [9] has a significant improvement over LIC and HCC algorithm s as for as the cost of cluster mainte nance is consider. Most of protocols execute s the clustering proced ure periodicall y, and re-cluster the nodes from time to time in order to satisfy so me specific characteristic of clusterheads. In HCC, the clustering scheme is performed periodically to check the “local highest node degree” aspect of a cluste rhead. When a clusterhead finds a member node with a higher degr ee, it is fo rced to hand over its clusterhead role. Th is mechanism, involves frequen t re- clustering. In LCC the c lustering algorith m is divided into tw o steps: cluster formation and cluster maintenance. The cluster formation sim ply follows LIC, i.e. initi ally mobile nodes wi th the lowest I D in their ne ighborhoods are chosen as clusterheads. Re-clustering is event-driv en and invok ed in only two cases: • When two clusterheads move into the reach range of each other, one gives up the clusterhead role. • When a mobile node canno t access any clusterhead, it rebuilds the cluster structur e for the network according to LIC. Hence, LCC significantly impr oves cluster stability by relinquishing the requ irement that a clusterhead should HAVE some special features in its lo cal area. But the second case of re-clustering in LCC indicates that a si ngle node’s movement may still inv oke the complet e cluster structure recomputation and thus results in larg e communication overhead. 2.4.2 Adaptive cl ustering for mobile wireless network [10]. ensures smal l communi cation overhead for bu ilding clusters because each mobile node broadcasts only one message for the cluster co nstruction. In th is adaptive clusterin g scheme, every m obile node i keep it s own ID and the ID of its direct neighbors i n a set Gi. Each mobile node wi th the lowest ID in their local area declares to be a clusterhead and set its own ID as its cluster ID (CID). The CID information includes a mobile node’s ID and CID. When a mobile node i receives CID information from a neighbor j, it deletes j from its set Gi. If the CID information from j is a clus terhead claim, the mobile node ISSN : 0975-3397 Ratish Agarwal et al / Internat ional Jo urnal on Com puter Science and Engineering Vol.1(2), 2009, 98-104 101 checks its own CID aspect. If its CID is unspecified (it is not involved in any cl uster yet) or larger than the ID (CID) of j, it sets j as its clusterhead. Th e process continues till all mobile nodes access some cluster. After cluster formation is completed, clusterheads are no longer us ed in any further cluster maintenance phase. In th e maintenance phase, when a mobile node i fi nds out that the distance between i tself and some node j in the sam e cluster becomes greater than 2-h op, it invokes a cluster maintenance m echanism. If node i is a direct neighbor of the node with the highest intra-cluster connecti vity in its cluster, it remains in the clu ster and removes node j; otherwise, it joi ns a neighboring cluster. As soon as there is no proper cluster to join, it forms a new cluster to cover itself. Since this mechanis m likely forms new clusters but without any cluster elimination or merg e mechanisms, the cluster size decreases and the number of cl usters increases as time advances. Eventually , almost every m obile node forms a single-node clust er, and the cluster struct ure disappears. 2.4.3 3-hop b etween adjace nt clusterheads (3-hBAC) [11]. This algorithm introduce a new node st atus, “clusterguest”, which m eans this node is not within the transmission range of any clusterheads, but w ithin the transmis sion range of s ome clusterm embers. The cluste r formation alway s begins from the neighbo rhood of the m obile node with the lowest ID (assuming it is m obile node mo) in a MANET. The mobi le node with the highe st node degree in mo’s closed ne ighbor set is chosen t o be the first cl usterhead. All the direct neighbors of this clusterhead change status t o “clustermember.” After th e completion of th e first cluster, the cluster formation procedure can be performed in paral lel in the network. A clusterm ember or a direct neighbo r of any clustermember wi th status “unspecified” (i ndicating it is not included in any cluster yet) are denied serving as a clusterhead. A mobile node, which is n ot denied clusterhead cap ability, declares as a new clusterhead wh en it is with t he highest node degree in its neighborh ood. When a mobile node fi nds out that it cannot serve as a clusterhead or j oin a cluster as a clustermem ber, but some nei ghbor is a clustermem ber of some cluster, it joins the corresp onding cluster as a cl usterguest. For cluster maintenance, this algorithm keeps the adjacent clusterheads at least two-hops aw ay. So when two clusterhe ads move into the reach range of each other, one is required to give up its clusterhead role. W ith the help of clusterguest, 3hBAC does not raise ripple effect when re-clustering, which means the clusterhead re-election wil l have no affect on the status of mobile nodes out side these two clusters. For anot her case, when a mobile node m oves out of the ranges of al l clusters, it can join a cluster as a cluste rguest if it can reach some clustermember(s) of that cluster. Hence, there is no need to form new clusters i n order to cover such a single node as in LCC and the cluster topology does not change. This can reduce the number of cluste rs and eliminate sm all unnecessary clusters. 2.4.4 Passive clusteri ng [12]. Most of the clusteri ng algorithms requ ire all the mobile n odes to announce clus ter- dependent in formation repeatedly to build and maintain th e cluster structure, and th us clustering is one of the m ain sources of control overhead. A clust ering protocol that does not use dedicated control packets or signal s for clustering specific decision is called Passive Cl ustering. In this schem e, a mobile node can be in one of th e following four states: initial, clusterhead, gateway , and ordinary node. Al l the mobile nodes are with ‘initial’ state at the be ginning. Only m obile nodes with “initial” state have the potentia l to be clusterheads. When a potential clu sterhead with “initial” state has something to send, such as a flood search, it declares itself as a clusterhead by piggybacking its state i n the packet. Neighbors can gain knowledge of the clust erhead claim by monitoring t he “cluster state” in the packet, and then r ecord the Cluster head ID and the packet receiving time. A mobi le node that receives a claim from just one cluste rhead becomes an ordinary node, and a mobile node that hears more cl aims becomes a gateway. Since passive clustering does not send an y explicit clustering -related message to maintain the clus ter structure, each node is responsible for updati ng its own cluster stat us by keeping a timer. When an ordinary node does not receive any packet from its clusterhead for a given perio d, its status reverts to “initial”. 2.5 Power-aware clustering 2.5.1 Load balancing clust ering (LBC) [13] prov ide a nearby balance of l oad on th e elected clusterheads. Once a node is elected a clusterhead it is desirable for it to stay as a clusterhead up to some maximum specified amount of time, or budget. The budget is a user de fined restriction placed on the algorithm and can be modi fied to meet the unique characteristics of the system, i.e ., the battery life of individual nodes. In this algorithm each mobile node has a variable , virtual ID (VID), and t he value of VID is set as its ID num ber at first. Initially, m obile nodes with the highest IDs i n their local area win the clusterhead role. LBC limits the maxi mum time units that a node can serv e as a clusterhead continuously, so when a clusterhead exhau sts its duration budget, it resets its VID to 0 and becomes a non-clusterhead node. When two clusterheads move into the reach range, the one with higher VID wins the clusterhead role. when a clusterhead resigns, a non-clusterhead with the largest VI D value in th e neighborhood can resume the clust erhead function. The newl y chosen mobile n ode is the one w hose prev ious total clusterh ead serving time i s the shortest in its nei ghborhood, and this should guarantee good energy l evel for being a new clusterhead. However, the drawback is that the clusterhead serving ti me alone may not be a good indi cator of energy consum ption of a mobile node. 2.5.2 Power-aware conn ected dominant set [14] is an energy-efficient clustering scheme which decreases the size of a dominating set (DS) witho ut impairing its fun ction. The unnecessary mobile nodes ar e excluded from the domi nating set saving their energy consumed for serving as clusterheads. Mobile nodes in side a DS cons ume more battery energy than those outside a DS because mobi le nodes inside the DS bear extra tasks, includi ng routing i nformation updat e and data packet relay. Hence, it is nece ssary to minimize the energy consumption of a D S. In this scheme Energy level (el) in stead of ID or node degree is used to determi ne whether a node should serve as clusterhead. A mobile node can be deleted from the DS when its close neighbor set is covered by one or two dominatin g neighb ors, and at the same ti me it has less residual energy than the dominating neighbors. This sch eme ISSN : 0975-3397 Ratish Agarwal et al / Internat ional Jo urnal on Com puter Science and Engineering Vol.1(2), 2009, 98-104 102 cannot balance the great di fference of energy consum ption between dominating no des (clusterheads) and non-domi nating nodes (ordinary nodes) because its objective is to minim ize the DS rather than to balance t he energy consum ption among all mobile nodes. Hence, mobile nodes in the DS still likely deplete their energy at a much faster rate. 2.5.3 Clustering for energy conservation [15] assumes two node types: master and sl ave. A slave node must be connected to only one master node, an d a direct connection between slave nodes is not allowed. Each m aster node can establish a cluster based on connecti ons to slave nodes. The area of a cluster is determined by the farthest distance between the master node and a slave node in the cluster. Master nodes are s selected in advance, and can only serve a lim ited number of slave nodes. The purpose of of this schem e is to minimize the transmi ssion energy cons umption summ ed by all master- slave pairs and to serve as many slaves as possible in order to operate the network wi th longer lifet ime and bette r performance. Two schem es, single-phase clustering an d double-phase clustering, are prop osed in [15]. In single-ph ase clustering, in itially every master node w ill page slave nod es with the allowed maximum energy. For each slave that receives one or multipl e paging signals, i t always sends an acknowledgment message back to the m aster from which i t receives the strongest paging si gnal. Since a master node can serve only a limited number of slav es, it first allocates channels for slaves that only receive a singl e paging signal from itself. If any free channels remain, other slave nodes, which receive more than one paging si gnal, are allocate d channels in the order of the power level of the paging signal recei ved from the master node. For those slave no des, which do not receive a channel from a master in the channel allocation phase, are dropped in the furt her communication phas e. This mechanism can reduce the call drop rate by giving priorit y to those slave nodes that only receive single paging si gnals in channel allocation. Slave nodes, which r eceive multiple paging signals, always try to communicate w ith the nearest master. E ach connected master-slave pair communicates with the minimum transmis sion power in order t o save energy. To further lower the call drop rate, double-p hase clustering re-pages for slaves, which do not receive a channel in the first round, in its range. The channel allocation procedur e also follows the received signal strength. The drawb ack of this scheme are pagi ng process before each round of communication consumes a large amount of ener gy. Master node electi on is not adaptive, and the method of select ing the master node i s not specified. 2.6 Combined-weight based clustering 2.6.1 Weighted cl ustering algorithm (WCA) [16] selects a clusterhead according to the number of nodes it can handle, mobility , transmission po wer and batte ry power. To avoid communications o verhead, this algorithm is not pe riodic and the clusterhead election pr ocedure is only invoke d based on node mobility an d when the current do minant set is incapable to cover all the nodes. To ensure that clusterheads will no t be over-loaded a pre-defi ned threshold is used which indicat es the number of nodes each clusterhead can ideally support. WCA selects the clusterheads according to the weight value of each node. The weight associ ated to a node v is defi ned as: Wv = w1  v + w2 Dv +w3 Mv +w4 P v ------------------ -- ---- (1) The node with the minimum weight is selected as a clusterhead. The weighting factors are chosen so that w1 + w2 + w3 + w4 = 1. Mv is th e measure of mobilit y. It is taken by computing the running average sp eed o f every node duri ng a specified time T.  v is the degree difference.  v is obtained by first calculating the number of neighbors of each node. The result of this calculatio n is defi ned as the degree of a node v, dv. To ensure load balanci ng the degree difference  v is calculated as |dv -  | for eve ry node v, where  is a pre-defined threshold. The parameter Dv is defined as the sum of distances from a given node to all its neighbors. This fact or is related to energy consumption since more power is needed for l arger distance communications. The para meter Pv is the cumulative time of a node being a cluste rhead. Pv is a measure of how much battery power has been consumed. A clusterhead consumes more battery than an ordinary node because it has extra responsibilities. The clusterhead election algo rithm finishes o nce all the nodes become either a clusterhead or a m ember of a clusterhead. The distance between members of a clusterhead, must be less or eq ual to the transmissio n range betw een them. No two clusterheads can be immediate neighbors 2.6.2 Entropy- based Weighted clustering al gorithm [17] In WCA high m obility of nodes leads to high fre quency of reaffiliation which increase the network o verhead. High er reaffiliation frequency lead s to more recalculations o f the cluster assignment resu lting in increase in co mmunication overhead. Entropy base d clustering overcomes the drawbac k of WCA and forms a more stable net work. It uses an entropy- based model for evalu ating the route stability in ad hoc networks and electing clus terhead. Entropy pres ents uncertainty and is a measure of th e disorder in a system. So it is a better indicator of th e stability and mobilit y of the ad hoc network. 2.6.3 Vote-bas ed cl ustering algorithm [18] i s based on two factors, ne ighbors' num ber and remaining battery time of every mobile host (M H) Each MH has a uni que identifie r (ID) number, which is a po sitive integer. The basic information inside the netw ork is Hello message, w hich is transmitted in the common channel . Making use of node l ocation inform ation and power information , this algorith m introduce the co ncept of "vote". The Hello message form at is given below. MH_ID item is MH's own ID and CH_ID item is MH's c lusterhead ID, Vote item means MH 's vote value, i.e. we ighted sum of nu mber of valid neighb ors and remaining battery time. Op tion item is used to realize cluster load balance. MH_ID CH_ ID Vote Op tion Hello message form at Vote = w1 x (n/N) + w2 x (m/M) ----- ------------ ----------- (2) ISSN : 0975-3397 Ratish Agarwal et al / Internat ional Jo urnal on Com puter Science and Engineering Vol.1(2), 2009, 98-104 103 w1; w2: Weighted coefficien t of lo cation factors an d battery time, respectively, n: Number of n eighbors, N: Network size or the Maxi mum of mem bers in a cluster, m: Remaining battery time, M: The maximum of battery tim e remaining bat tery time. Each MH sends a Hello message random ly during a Hello cycle. If a MH is a new user to the netw ork, it reset "CH_ID " item. That means th e MH does not belong to any cluster and does not know whether it has neighbor hosts. Each MH counts how many Hello messages it can receive during a Hello pe riod, and considers the number of received Hello m essages as its own n. Each MH sends anothe r Hello message, in whi ch "vote" item is set to its ow n vote value and got from Equatio n. Recording Hello message during second Hello cycles, each MH knows the sender wi th highest vote and not belongs to any existing cluster is its cluster h ead. It set its next sending Hello message item "CH_ID" to the cl uster head's ID value. When two or more mobile nodes receive the same num ber of hello packets, the one wh o owns the lower ID will g et priority. Following this approach, every M H knows its cluster head ID after two Hello message periods. 2.6.4 Weight-based adap tive clustering algorith m (WBACA) [19] Drawback s of WCA alg orithm is that a ll the nodes in the network have to know t he weights of all the other nodes before starting the cl ustering process. This process can take a lot of t ime. Also, two cluster heads can be one-hop neighbors, whi ch results in the cluste rs not necessarily being spread out in the network The clust ering approach presented in WBACA is based on the avai lability of posi tion inform ation via a Global Positi oning System (GPS). T he WBACA considers following parameters of a node for clusterh ead selection: transm ission pow er, transmission rate , mobility, battery power and degree. Each node i s assigned a weight that indicates its suitabili ty for the clusterhead role. The node wi th the smallest weight is chosen as the clusterhead. The weight of a node N is defi ned as: WN = w1*M + w2*B + w3*Tx+ w4 *D+ w5 /TR ------(3) where wl, w2, w3, w4. and w5 are the weighing factors for the corresponding syst em parameters li sted below: - M: Mobility of the node B: Battery power Tx: Transmis sion power D: Degree difference, and TR: Transmission rate This algorithm further allo ws no two clusterheads to be one-hop neighbors of each othe r. Overlapping clusters are connected through Gat eways (nodes connect ing two clusterheads). All the ordinary nodes are one-hop from their clusterheads 2.6.5 Connectivity, energy & mobility driven Weighted clustering algorithm (CEMCA) [20] The electi on of the cluster head is based on th e combination of several sign ificant metrics such a s: the lowest node mobility, the highest n ode degree, the highest battery ener gy and the best t ransmission range. This algo rithm is completely distri buted and all nod es have the same chance to act as a cluster head. CEMCA is composed of two main stages. Th e first stage consists in the election of the cluster head and the second stage consis ts in the grouping of members in a cl uster. Normalized value of mobility, degree an d energy level is calcula ted and is used to find the quality (n ormalized to 1) for each node. The node broadcasts its quality to their neighbors in order to compare the better among them. After this, a n ode that has the best qu ality is chosen as a clusterhead. In the second stage the construction of the cluster members set is done. Each clusterhead defines its neighbors at two hops m aximum. These nodes fo rm the members of the cluster. Next, each cluster head stores all information abou t its members, and all nodes record th e clusterhead identifier. Th is exchange of informatio n allows the routing protocol to function in the clust er and between the clusters. III. C ONCLUSION We have reviewed several clusteri ng algorithm s which help organize mobile ad hoc net works in a hierarchical manner and presented their main characteristics. With th is survey we see that a cluster-based MANET ha s many im portant issues to examine, such as the clus ter structure stability, th e control overhead of cluster construction and m aintenance, the energy consumption of mobile no des with different cluster-related status, the traffic load distribu tion in clusters , and the fairness of serving as clusterheads f or a mobile node. IV. R EFERENCES [1] M. Gerl a and J. T. Tsa i, “Mu ltiuser, Mobile, Multimedia Radio Network,” Wire less Netwo rks , vol. 1, pp. 255–65, Oct. 1995 [2] A.D. Amis, R. Prakash, T .H.P Vuong, D.T. Huynh. "Max-Min D- Cluster Formation in Wireless Ad Hoc Networks". In proceedings of IEEE Conference on Computer C ommunications (INFOCOM) Vol. 1. pp. 32-41, 2000 [3] G. Chen, F. Nocetti, J. Gonzalez, and I. Stojmenovic, “Connectivity based k-hop clustering in wireless networks”. Proceedings of the 35th Annual Hawaii International Conference on System Scie nces . Vol. 7, pp. 188.3, 2002 [4] F. Li, S. Zhang, X. Wang, X. Xue, H. Shen, “Vote- Based Clustering Algorithm in Mobile Ad Hoc Ne tworks”, In proceedings of International Conference on Networking Technologies, 2004 [5] T. Ohta, S. Inoue, a nd Y. Kakuda, “An Adaptive Multihop Clustering Scheme for Highly Mobile Ad Hoc Networks,” in proceedings of 6th ISADS’03 , Apr. 2003 [6] I. Er and W. Seah. “Mobility-bas ed d-hop clustering algorithm for mobile ad hoc networks”. IEEE Wi reless Communications and Networking Conference Vol. 4. pp. 2359-23 64, 2004 [7] P. Basu, N. Khan, and T. D. C. L ittle, “ A Mobility Based Metric for Clustering in Mobile Ad Hoc Ne tworks,” in proceedings of IEEE ICDCSW’ 01 , pp. 413–18, Apr. 2001 [8] A. B. MaDonald and T. F. Znati, “A Mobility-based Frame Work for Adaptive Clustering in Wireless Ad Hoc Networks,” IEEE JSAC , vol. 17, pp. 1466–87, Aug. 1999 [9] C.-C. Chiang et al. , “Routing in Clustered Multihop, Mobile Wireless Networks with Fading Ch annel,” in proceedings of IEEE SICON’97 , 1997 [10] C. R. Lin and M. Gerla, “Adaptiv e Clustering for Mobile Wireless Networks,” IEEE JSAC , vol. 15, pp. 1265–75, Sept. 19 97 ISSN : 0975-3397 Ratish Agarwal et al / Internat ional Jo urnal on Com puter Science and Engineering Vol.1(2), 2009, 98-104 104 [11] J. Y. Yu and P. H. J. Chong, “3hBAC (3 -hop between Adjacent Clusterheads): a Novel Non-overla pping Clustering Algorithm for Mobile Ad Hoc Networks,” in proceedings of IEEE Pacrim’03 , vol. 1, pp. 318–21, Aug. 2003 [12] T. J. Kwon et al. , “Efficient Flooding with Passive Cluster ing an Overhead-Free Selective Forward Mechanism for Ad Hoc/Sensor Networks,” in proceedings of IEEE , vol. 91, no. 8, pp . 1210–20, Aug. 2003 [13] A. D. Amis and R. Prakash, “Load-Balancing Clusters in Wireless Ad Hoc Networks,” in proceedings of 3rd IEEE ASSET’ 00 , pp. 25–32 Mar. 2000 [14] J. Wu et al. , “On Calculating Power-Aware Connected Dominating Sets for Efficient Routing in Ad Hoc Wireless Networ ks,” J. Commun. and Networks , vol. 4, no. 1, pp. 59–70 Mar. 2002 [15] J.-H. Ryu, S. Song, and D.-H. Cho, “New Cluster ing Schemes for Energy Conservation in Two-T iered Mobile Ad-Hoc Networks,” in proceedings of IEEE ICC’01 , vo1. 3, pp. 862–66, June 2001 [16] M. Chatterjee, S. K. Das, and D. Turgut, “An On-Dem and Weighted Clustering Algorithm (WCA) for Ad hoc Networks,” in proceedings of IEEE Globecom’00 , pp. 1697–701, 2000 [17] Yu-Xuan Wang, Forrest Sheng Bao, "An Entropy-Based Weighted Clustering Algorithm and Its Optim ization for Ad Hoc Networks," wimob,pp.56, Third IEEE International Conference on Wireless and Mobile Computing, Networking and Communications ( WiMob 2007), 2007 [18] F. Li, S. Zhang, X. Wang, X. Xue, H. Shen, “Vote- Based Clustering Algorithm in Mobile Ad Hoc Networks ”, proceedings of International Conference on Networking Technologies, 2004 [19] S.K. Dhurandher and G.V. Singh” Weight-based adaptive clustering in wireless ad hoc networks” IEEE 2005 [20] F.D.Tolba, D. Magoni and P. Lorenz “ Connectivity, energy & mobility driven Weighted clusteri ng algorithm ” in proceedings of IEEE GLOBECOM 2007 ISSN : 0975-3397