CC BY
0Sensors 2010, 10, 5233-5262; doi:10 3390/s100505233
O PEN ACCESS
SSnSDIS
ISSN 1424-8220 www m dpicom /pumal/sensors
Article
NEURON : Enabling Autonom icity in W ireless Sensor Networks
12 1 2
Anastasias Zafeiropoulos ''*, PanagiotisGouvas , Athanassios Liakopoulos ,
Gregoris Mentzas 1 and Nikolas M itrou 1
1 N atonal TeChnicadUniersity of Athens, Heroon Polytexneiou, 15773, Zograibu, Gieege;
E-M ails: pgouvas@ m ailntuagr PG .) ; gm entzzas@ m ailntuagr (G M .) ; m itrou@ csntuagr (NM.)
Greek Research and Technology Network:, Av. M esogron 56, 11527, Athens, Greece;
E-M ail: alako@ grnetgr
* Authorto whom correspondence should be addressed; E-M ail: tzafeit® cnntuagr; Tel: +30-210-7474255; Fax: +30-210-7474490.
Received: 1 February 2010; in revised form : 1 April 2010 /Accepted: 14 M ay 2010 / Published: 25 M ay 2010
Abstract: Future W ireless Sensor Networks W SNs) will be ubiquitous, large-scale networks interconnected with the existing IP infrastructure. Autonom ic functionalities have to be designed in order to reduce the com plexitty of their operation and m anagem ent, and support the dissem ination of know ledge within a W SN . In this paper a novel protocol for energy efficient deploym ent, clustering and routing in W SNs is proposed that focuses on the incorporation of autonom ic functionalities in the existing approaches. The design of the protocol facilitates the design of innovative applications and services that are based on overlay topologies created through cooperation am ong the sensor nodes.
Keywords: wireless sensor network; autonomicity; energy efficiency; clustering; hierarchical routing; p2p; overlay; NEURON
1. introduction
In the last decade, there has been a great evolution in the sensor networking world. A vast am ount of sm all, inexpensive, energy-efficient, and reliable sensors witthwireless networking capabilities is available worldwide, increasing the number of sensor network deployments [1,2]. The adoption of IPv6, combined with the advanced networking capabilities of modern sensor nodes, enables the
integration of sensor networks into the existing IP networking infrastructure [3]. IPv6 provides a huge address space for networking purposes, while concurrently leading to the rapid developmentof many useful applications. Follow ing these advancem ents, the vision of a W orld-W ide Sensor W eb [4] is becom ing a reality, since the trend in next generation networks is to create fully interconnected infrastructures, consisting of m any heterogeneous networks.
The transition to large scale wirele^ sen^r networks (W SNs) increases the complexity in their operationand m anagem ent and raises stability andscalability isaies [5]. This is due to the special requiem ents that are im posed from the sensor nodes and the characteristics of the W SN s. B attery lim iatons and the need for lifetim e m axim izaton m ake necessary the overall pow er consum pton m inim ization during operation. Energy-efficient operation has to be realized taking in account the existence of heterogeneous sensornodes with diverse com puting and storage capabilities. Furtherm ore, the highly volatile and dynam ic nature of W SN topologies—due to continuous nodes joins and leaves orm obility in the W SN field—requires the continuous intervention of the network adm inistrator.
Sef-organization m echanism s are proposed in order to reduce the com plexity in the operation and m anagem ent of W SN s. Considering the specific characteristics of arch networks, the m anagem ent of sensor infrastructures should be as autonom ic as possible, ie., the w irele^ sen^rs should m anage themselves with minimum or no human intervention [6]. Autonomic functionalities have to be developed tow ards the sensor network evolution focusing on the introduction of functionality that will enable the provision of advanced services with low pow er consum ption. Establidm ent of cooperation am ong nodes, ability to retrieve the current status of each sensor node, ubiquitous access to data and optim isation of the overall sensor network perform ance are exam plies of functionalities that facilitate the efficient operation of sensor nodes from the energy and perform ance perspective. By developing pelf-f incthnalities, autonom ic m onitoring will be realized, adaption to environm ental changes will be supported and operational and m anagem ent chstwi]l be reduced.
A prom ising m ethod tto enable autonom icity in wirele^ heterogeneous environm ents is the adoption of decentralized schemes [5,7]. Centralized componentsm ake the system vulnerable in the sense that they are single points of failure. Furtherm ore the entire traffic load m ay be needed tto be directed towards them , which is energy consuming, especially in dense networks [8,9]. On the contrary, decentralized ahem es are fully salable and achieve hom ogeneous load distribution. Current research efforts show that decentralization m ay be achieved through the exploitation of techniques that are applied in peer-to-peer p>2p) networks since relevant protocols rely on decentralized structures, arch as D istributed H a±L Tables DHTs) [10]. These techniques are basedon the creationof an overlay topology and the im plem entation of specific m echanism s over it [4]. P2p overlays may be form ulated over sensor networks tto elim inate the need for proxy support, tto provide efficient data lookup, tto limit broadcasts and tto enable flexible access tto the sensed data. However, the m aintenance of an overlay topology is challenging in dynam ic environm ents where nodes join or leave the network [11,12] and requires the exchange of a large num berof m essages am ong the participating nodes [13].
The transition to an overlay approach for the W SN deploym ent, operation and m anagem ent has tto be combined with energy efficient. techniques, such as clustering and hierarchical routing techniques [14]. Nodes that are elected tto operate as Cluster Heads (CHs) are involved mainly in routing and packet relaying functions while the rest nodes perform a m inim um ^t of operations [15].
Furthern ore, the availablB rou^tiig inform atonm ay be exploilBd from upper ]&yer m echani^ s and make them more effi-iient-
In tthis paper we propose NEURON, a novel prottoool for au^ttonomic deploym ent distEr forn uation and hierarchicaL routing in wileless sensor nettwolks. NEURON com bines the oonoepts of sef-orjanization, deoentraliza^tion and optimization in building networking infrastricture. The oom bination of these conc^ts is auoialL, since it perm its autonom ic deploym ent and :e-CDnfigu:atDn of the network and suppcrts the developm ent of sEf-optim isation functionallities. Energy efficii^icy is achieved and networklifetim e is atended th:Dugh cJustering and ^i^aarch^ical routing m echanLisn s. NEURON does not m ake any assum ptions for the locationof the sensors, their capabiliti^ and the roles that m ay underake. It is designed in order tto be stable and scaiablB and to faoilitatв the cr^tion of overlay topologies in the sensor network by exploiting routing inform ation stored in the nodes.
The paper is organized as foUow s. S^ction two briefly presents the relatBd work on the fBld while sectionthree describes indetail the proposed prottocol andits m echanism s. SeGtDn:fDu: detaills how NEURON may facilítate the autonom ic provision of ad^anced services in a W SN . S^ction fiive discuses the smuation resulsand section six concluides the paperwith a short summaty of ourwortk and a presentation of open researh issues and future work.
2. B aokglDund—RBlated W ork
21. Autonom ic ChalaGtв:ist±:s and IP ]htвglatDn
Since a com m on tBchnology is benefcial, the trend in the design and im plem entation of future nettwolks is the intвg:atDn of sensors with the existing IP netwolks. W SNs are usually considered as low-power, wireless personal area nettwolks (LoW PANs). LoW PANs are poised to forn the next tBr of the IntErnet by cDnnecting billion of sensor nodes to the IntErnet and using the existing com puting infestructure [3,16]. ExtBnding IP to LoW PANs was once considered imprácttical [3] due to the rsourB-intEnsive ovelhead im posed by IP com m uniGatDn to enerзy-cDnS:ained devices. In order to handle these issues, 6LoW PAN was int^DduGed as an adapta^tion layer that enables efficient IPv6 com m uniGatDn over80215.4 llinks [3].
The tran^tion to IP is Gaucial since it enables the design and deploym ent of autonom ic f]nct-Drali-iesin W SNs. IP netwolking ::ha:actвrist±s may be exploilBd by the network layerin order to be responsive and adaptive while remaining ene^gy-effiGient. The deploym ent of autDnomio fiunotiDna"itвs is helpful since the protocolls and the tв::hniques that are being developed form onitoling and m anagem ent of W SN s should have m inim al GDnfigu:atDn, preferably wortk "out of the box", be easy to bootstrap, andвrablв the netwo::ktD selfheal given the inherentunlBliable oha^aote:i¡=trio of these devices. The basic autonom ic fUnotDnalities that have to be suppored by sensor nodes are [17]:
• Sвf-GDn:fguration: change GDnfigu:atDnpa:am Beis (e^-g., data ao^i^tDn:э^t^) aG:G:D:^ing to the GDrd^itDr^s in the sensorenvi:tDr]m en^tt,
• Self-Dptimi^t:Dn: fine-tune f^ch sen^sD:en^t:ty in orderto achieve pr^-detBam ined goal^, while low en^gy GDn^sumptDn and optimal quality of ^l^iGB have to be guarantBed based on GD:l^tive actions,
• Self-healing: detect system m alfUnathns or failures and start corrective actons based on defined policies and cooperation am ong neighbouring nodes to recover the network ora node,
• Self-protection: recognize po^ible source of problem s or threats in the sensor network and take proactive m easues,
• Self-awareness: ^n^ network changes in the sensor environm ent and be aw are of the capabilities and the status of your neighbours.
Self-organizing system s com pletely rely on localized decision proce^es and nodes have to follow three basic m etthods inorder to im plem ent the desired behaviour; interact w ithother nodes intheir neighbourhood, adapt their local state according to the conditions in their environm ent, and include probabilistic tEchniques in their decisions [5,18]. Based on the acquired know ledge, it is desirable that the initial sensor network configuration takes advantage of the underlying physical arising and topological characteristics so as to a^ign responsibilities to nodes that are best aiited to perform certain sensor network duties [19]. This preliminary assignment of duties facilitates future reorganization of the network in order to easily adapt to any changes.
Taking into account these considerations, NEURON's mechanisms are designed to operate in a fully autonom ic manner:. The network entities cooperate and exchange information based on the policies that are applied by these mechaniOT s.Thus, the sensor network may be эe]f-chnfigurEd and ^lf-optim iffid, it supports ^If-aw arene^ through proper di^am ination of inform ationand is able to reacttto possible failures.
2 2. Clustering and Routing Algorithms for W SNs
Since energy efficiency is a crucial characteristic fortthe network lifetim e in W SN s, it is necessary to design m echanii s that are not energy consum ing. Several techniques have been proposed recently for energy efficient- clustering and routing in W SN s. It has been proved that by using hierarchical (tiered) architectures, the network lifetime can be extended significantly [20,21]. A hierarchy may be created by using clustering m echaniOT s and hierarchical routing has to be selected.
Clustering has been proven to be energy-efficient since data routing and relaying are only operated by cluster heads (CH s) [22]. CH s proce^, filter and aggregate data sent by cluster m em bers, thus reducing network load and bandw idth utilisation. Besides, non-CH sensors are not involved in routing and relaying data and transm issions are only operated by CH s [15]. During the clustering process, it is necessary to take into account aspects arch as the cluster size and form ation, criteria forCH s election, how to control inter-cluster and intra-cluster collisions and energy saving issues [15]. For the proper design of a clustering m echanism , multple challenges have to be addre^ed. Clustering has to be efficient in term s of processing com plexity and m essage exchanges and the clustering form ulation technique has to be autonom ic, ie. each node takes decisions independently of the other nodes. The clustering process has to be com pleted w itthina boundednum ber of iterations and, after the clusters have been form ulated, each node operates either as a CH or as a sim ple node. Furtherm ore, adequate distribution of CH s overthe sensor field has to be accom plished [4]. Finally, since CH s consum e m ore energy in aggregating and routing data, it is es^ntial to have an energy-efficientm echanii for CH s election and rotation [15].
Hierarchical routing protocols have to be correlated with the clustering form ulation m echanism s in order to reduce the routing m essages exchanged and, thus, decrease energy consum pton [14]. These protocols ate closely related. to the clustering mechanism and usually support multi-hop communication, where each clustermember forwards its data to the CH and the CH s forwards the data to the gateways in the network:. Since many of the proposed system s involve large networks, it is essential to provide routing infrastructures that concurrently offer small routing state and robustness [23,24]. Limiting routing states stored in the nodes is crucial for scalability and efficiency [25] while robustness entails handling efficiently topology and connectivity changes due to node failures and envionm entalim pact. M oreover, it is im portantto provide m echanism s that are able to handle heterogeneous nodes with diverse capabilities in the network [26].
Energy-effcient clustering and routing algorithm s forwieless sensor networks are presented in the literature [27-29]. Cluster formation is typically based on the energy reserve of sensors and sensor's proxim ity to the CH [30]. Low -Energy Adaptive Clustering Hierarchy LEACH ) [15] was one of the first clustering algorithm s proposed for sensor networks. LEACH is a distributed, proactive, dynam ic algorithm that form s clusters of sensors based on the received signal strengtthand uses local CH s as routers to the sink:. It includes random ized rotation of the high-energy cluster-head position, ie., delegate cluster-head functionality am ong the various sensors in order to preserve the battery of a single sensor. LEACH clustering terminates in a constant number of iterations but it does not guarantee good CH distribution and assum es uniform energy consum ption for CH s. On the contrary, HEED [31] makes no assumptions on energy consumption and achieves to selectwell-distributed CHs. However, it prerequisites that the nodes in the network are quasi-stationary and have equal capabilities. ED AC [32] is an improvement of LEACH for heterogeneous environments that supports energy-driven CH s rotation and extends the node lifetm es, while LENO [33] proposes a dynam ic CH rotation algorithm that outperform s both LEACH and EDAC. PEGASE [34], LBCS [9] and EECS [35] are various enhancem ents of LEACH, where energy dissipation isbalanced among sensor nodes under special conditions or assumptions. PEGASE and LBCS assum e that all nodes have location information about all other nodes, while EECS supposes that all nodes are stationary and uniform ly dispersed within a sensor field. In all the above cases, the sensors are directly connected to theirCHs. Furthermore, layered approaches exist such as SOHS [36] and EECF [15] in which sensors are organized into clusters creating a hierarchical topology and HM PR [37] where the W SN isinitialy constructed as a layered network. These protocols necessitate the existence of fixed centralized nodes or fixed lifetm es and identifiers for the sensor nodes. Finally, TinyH op [38] is proposed as a reactive routing protocol that aim s to m inim ize the num ber of m essages necessary to perform routing. Therefore, this protocol avoids the energy consum ing periodic beacon m essages generated by other proactive routing protocols. TinyH op m ay also be configured to lim it the flooding of m essages within the scope of the cluster. In our approach, the routing am ong the CHs is designed in a reactive m anner, taking into consideration existing reactive protocols that are already designed for m obile ad hoc networks [39] such asDSR [40].
The NEURON protocol is designed as a new approach forproviding autonom ic and energy efficient clustering and routing in W SNs without assum ing specific pre-conditions for the nodes functionality and their knowledge. As described in detail in Section 3, NEURON does not make any assum ptions about the locationof the sensors, theirheterogeneity, their capabilities, the pre-definitionof CH s or
centralized nodes and their unique identifiers. This is achieved since the clustering and routing m echanii s are ^lf-orga^izedandaie basedon the aw areness of the neighbouring environm ent of each node (self-aw areness). Furtherm ore, in order to m axim is network lifetim e, the ^lf-organization phase is designed in order to be short and energy efficient, characteristic that is not adequately addre^d from the existing protocols [27].
23. Overlay Networking in WSNs and DHTs
Autonom ic functionalities in sensor networks m ay be facilitated by the creation of virtual overlay topologies [41]. The creation of p2p overlay topologies was proposed in order to treat the underlying heterogeneous W SNs as a single unified network, in which interacting sensors can exchange inform ation (store and retrieve data) without considering the details of the infrastructure underneath. P2p protocols are selected since they rely on decentralized algorithm s, arch as Distributed Hash Tables DHTs). P2p overlays over (traditional) sensor networks elm inate the need for centralized proxies and provide easy publication and search for the sensed data [4]. They provide efficient data lookup, guarantees on lookup times and location independence. The overhead of building applications is distributed amongst pariipating nodes with no central point of failures and without global broadcasts [10,42,43].
Several approaches have been proposed for p2p overlay networking in sensor networks. In [44], a DHT -based ^rvice diaxovery protocol that constructs topology- aw are overlay networks in Harge-aale W SNs is proposed. In [4], a Chord-based P2P protocol, called Tiered Chord (TChord) that can ^am lesslyintegrate sensor networks w itthlP networks is descrbedin detail, m [45] ,CSN , a novel DHT based network protocol for ^nsor networks is proposed where bounded tim es for data lookup -in the order of O (ogN) m essages- m ay be achieved in an energy efficient manner. CSN follow s a hierarchical clustering approach where each cluster is form ed ina logical ring. CSN m akes system lifetime of the sensor network proportional to its effective use and scales well to Harge-aale sensor networks.
However, a generic m apping of DHT based protocols to sensor networks is considered challenging for various reasons [45]. These protocols interconnect nodes independently of their physical location and are not able to handle dynam ic changes in the sensor network topology. In addition, they prerequisite the m aintenance of routing inform ation am ong all nodes and require unique identifiers for each node. M ost of these challenges m ay be addres^d given the existence of an overlay topology formulation and maintenance mechanism and its cooperation with the existing routing and clustering m echanism s.
Several algorithm s have been proposed for overlay topology form ulation and m aintenance that range from gossping techniques [46] to exhaustive techniques [47]. The common characteristic of these mechanii sis that they presum e guaranteed communication among the network nodes. However their principles are divert. G o^iping techniques attem pt to identify the relative position of one node in the overlay topology by consulting adjacent nodes. Alternatively, exhaustive techniques attem pt to paffi through all nodes periodically in order to identify their relative position in the overlay topology. The topology form ulation m echanism has to cooperate with the routing protocol due to the m entioned requisite tthatany two nodes be able to com m unicate.
In our approach, as described in detail in Section 4, we propose the exploitation of routing information that is maintained by NEURON via the overlay topology formulation mechanism . In NEURON, routing information is updated continuously in each node's routing cache and may be consulted from upper layer m echanism s. Given the existence and m aintenance of the overlay network:, DHT functionality may be applied—using the 6LowPAN address of each sensor as its unique identifier—and advanced services m ay be provided in an autonom ous m anner.
3. Proposed Protocol
In this section we present the Network basEd aUtonomic clusteRing and rOutiNg (NEURON) protocol, aim ing to address challenges that are accrued from the current trends in the W SNs evolution and the existing approaches for efficient deploym ent, operation and m aintenance of large in scale wireless sensor networks. NEURON is designed in orderto achieve the follow ing goals:
• Support autonom ic functionalities in a W SN : sensors deployed in the sensor field, establish independently communication with ttheirneighbours, become part of the W SN and gain access to the provided data and services without hum an intervention.
• Self-organize the nodes into clusters: clusters are self-formulated according to the current conditions in the network:, while cluster heads rotation extends the network liffetim e.
• Achieve fault tolerance and avoid dependence from specific purpose nodes: clusters are formulated autom atically in case of failures, since each node may be elected as CH .
• Reduce energy consum ption through hierarchical routing based on an overlay am ong the CHs: inform ation about the current CH s is dissem inated am ong the W SN and routing is provided through them .
• Achieve scalability by taking advantage of autonom icitty, decentralization and probabilistic techniques: there is no perform ance degradation as the size of the sensor network increases.
It is important to note that NEURON does not make any assum prions for its operation. It may be applicable to heterogeneous environm ents, in which nodes witth diverse computational and storage capabilities may be present. It does not require prior know ledge about location basedinform ationor the pre-assignm ent of specific roles in the sensor nodes. Its operationis basedon the know ledge of network based param eters within the W SN that could be estim ated in an autonom ous m anner by using an Autonomic Estimation Algorithm , as described in Section 31. NEURON may be effectively applied instatic and dynam ic topologies w here nodes continuouslyjoinor leave the networkor they are to som e extentm obile. Furtherm ore, its clustering and routing m echanism s—as presented in Sections 3 2 and 3 3—m ake it energy efficient since they suppress the num ber of m essages that are required for its operation and they supportm ulti-hop com m unication.
31. Autonom ic Estimation Algorithm
The know ledge of network-based param eters is crucial for achieving high efficiency and optim ising m echanism s in a W SN . The estim ation of such param eters is challenging in autonom ic environm ents, especially when there is no inform ation available at the initial deploym ent or after a topology change.
For example, in case of clustering, the knowledge of the network density (ie., the average number of single-hop neighbours of each sensor w itthinthe W SN) m ay facilitate the seltectionof the Hops-To-Live (HTL) param eter tor flooding lim itationwitthinthe W SN . W hen the densityof the networkis high the HTL param eter should be sm all and vice versa. Therefore, the know ledge of network-based param etters may im pact the efficiency of the m ecchanii s applied in divert W SN environm ents.
NEURON adapts its m echanism s basedonnetwork-wide param etters w itthout presum ing a priori knowledge of their values. An estimation of the parameters' values is autonomously produced and updated regularly w ithout im posing significant network overhead in term s of m essages exchanged. The m echaniOT is activated in each sensor when the network bootstraps or the topology changes significantly. It has to converge in a short number of cyc^ and to be applicable to large scale W SN s. An autonom ic m echanism is presented in this paper based on the principles of neighbour-based go^iping and specifically using averaging techniques based on neighbour-based gossiping [36].
Table 1. Averaging through Neghbour-go^iping.
converged=false; neighbconverged=false; do forever( if (cycle mod resetcycle ==0){ converged=false; neighbconverged=false; paramvalue = CountNeighbors(); ReceivedMSG=0; } if (ReceivedMSG!=0){ oldparamvalue= paramvalue; paramvalue=(paramvalue) / ReceivedMSG; if (Abs(oldparamvalue-paamvalue)<threshold){ converged=true; } } if (neighbconverged==false){ for (i=0;i<Neighbors.size();i++){ Send Frame[paramvalue,coverged] to Nodei } neighbconverged=true; } } On Message Receive Event{ paramvalue+=ReceivedFrame[paramvalue]; if (ReceivedFrame[converged]==false) neighbconverged=false; ReceivedMSG++; }
active thread passive thread
In NEURON , each node interacts witth its neighbours in order to calculate the mean value of a param eter Table 1). Each node calculates its initial value for a param eter and sends this value to its neighbours. In parallel each node receives from its neighbours their calculation about this param eter. After each 'cycle' of mutual exchanges, each node revises its calculation using a weighting average:
Value fromNeighbourU-----h Value fromNeighbourn)
U pdatedValue=-=-=—--=-=-=—--=—
W hen the updated value after the com pleton of a cyc^ diff^ le^ than a goecified threshold from the previous one, the param eter is considered as chnvErged on this node. In this case, the node, in the next cyc^, sends its converged value along w itth a flag that indicates that the node considers the parameter-estnmation as precise. Only when all me^ages that are received during a m essage-Exchange-ayc]e contain this convergence flag, a node decides to stop broadcasting its current value about a param eter. This procedure is repeated periodically in order to calculate the updated values of the network param eters.
This technique presents many advantages since there are no preconditions during the network bootstrapping, the estim ation is conducted in an ad hoc m anner and the algorithm converges quickly, even for large scale networks, while communication overhead is kept low . The frequency of the periodic estim ation m echanism is related to the application dynam icity. In NEURON , this technique is used for size and average density estim ation. The know ledge of these param eters is indicative for the possible network topology scheme and facilitates the good distribution of cluster heads in the clustering m echaniOT , as w e explainindetail in Section3 2. H ow ever, the sam e technique m ay be used for the auttonom ic calculation of other param eters within the W SN (eg, variance in the cluster sizes, available energy percentage).
For the density estim ation, each sensor node calculates the num ber of its neighbours and thus the converged param eter is chnsi(erEd to be the average network density. For the size estim ation, one or m ore predefined nodes in the network initialize the param eterNetwork_Size to 1 at step 0) while the rest nodes initialize the param eter Network_Size to 0. W hen the averaging protocol converges, the estim ated value is 1/(N ■ k) where N is the network size and k the num ber of nodes that initialized the param eter Network_Size to 1 [48]. The following adaptation is proposed in NEURON in order to estim ate the Network_Size autonom ically (without the need to predefine specific nodes that initialize the param eter Network_Size to 1). A ll nodes have a random num ber generator and expre^ their initiative to initialize their Network_Size param eter to1 w itth a certain probability. W e addre^ this probability as Pinit. The critical part of the adaptation is that all nodes respect the sam e probability. Pinit varies from 01 to 0 3. W hen the Network_Size param eter converges the converged value is approximatelly1/(N • Pinit).By inversing the chnvergedva]ue, an approximationof the networksize is available.
However, this adaptationprovides a parameter's estimationwithsome variance. In case that we desire to have more precise estimations, each node that chooses to initialize its Network_Size param eter to 1 has to accompany the broadcasted message with an additional field called SolicitatedG roup, m this fieldthe MAC address of the node is placed. Each node m aintains a cache that contains all the aolicitated M A Cs and in parallel, during each exchange of m essages, solicits the contents of its cache. Then, inadditiontto the convergence criteria that were formulated previously,
each node is not meant to be converged if the number of M ACs that exist in its cache is not equal to the num ber of M A C s that are sol •cited by its neighbours. Follow ing this adaptation the converged value is exactly (not approximately) [1/(N • NoM acs)] whala NoM acs stands for the number of elicited M ACs. The advantage of this adaptation is that i-generates extremely precise results, albeit at the expense of a langer am ount of m essages exchanged until convergence is achieved.
3 2. Cluster FolmuLaeon, M aineeпance and Update
The cluster formulation mechanim in NEURON has a significant impact on the W SN from m ultiple per^ectives. H im proves the energy efficiency and facilitates the design and deploym ent of higher layer protocols and applications. Clusters ate autonomous]yfornu]atad, maintained and updated basadonne;•ghbour to neighbour com m unicatbn am ong the sensor nodes. The Autonom ic EstimationAlgoritthm provides inform atbnneœssaty:fbr the optim isationof the CHs ^leationand distribution. Routing information acquired during the clustering process is stored in nodes' local caches and used by the routing álgorihm applied. C ontoled flooding is а!ю utilised in order to avoid traffic forwarding outside the cluster zone.
NEURON allows each node to become a CH according to specific arteria. A node that is œltected as CH acts as a proxy for the rest of the members in its cluster. Each node may be in two states; either belonging to a cluster or being in the procer of joining to a cluster. The clustering process starts only after the Autonom ic Estimation Algorithm has been converged and, thus, the size N and density d of the W SN is estimated. Based on these two parameters, each node decides to become a CH with a probability Palust- given by the follow ing equation :
Paust *KPI d)
available battery available memory , ,
where: KPI= 05*-=-- + 05*-=---2)
total_battery total_memoly
The Key Perform ance Indicator (KPI) in Equation 1 refers to the capabilities of each sensor. Nodes with better KPI present higher probability of becom ing CH s and rem aining in this status for a longer period of time untiltheir resources are reduced significantly, ü our case, the KPI is related with the available battery and m em ory of each node and is given in Equation 2. These param eters weie considered crucial for a sensor node deploym ent and operation in a W SN . How ever, any other parameterthaebetteraddresses application- goecific requiements (forthe KPI may be œlected.
According to equation 1, a sm aller num ber of CHs is expected to be elected in dense networks than in sparse ones. Inaddition, a higher num ber of CHs is anticipated in larger networks compared to m a Tlerones. Equation 1 was elected to aovera wide setof posible W SN topologies. However, if the network size ordensity is known a priori- m ore optim alpossbilhies Pclust m ay be selected.
The Pclust is updated regularly based on the current param eters N and d of the network:. This allow s NEURON to adapt to changes in the network topology or conditions and optim i the clustering form ulathon process. For exam ple, in case of a node transitioning from a CH operation m ode to norm al operationm ode, the nodes of the cluster decide to becom e CH s taking into cons;darltionthe latest estim atons of the parram eters N and d. m this case, if the average densityis reduced, m ole than one
CH may be elected. This proce^, combined with the cluster formulation update mechanic describEd in the next paragraph, facilitates the betterdistrjbuthn of CHs am ong the W SN and the extension in the network lfstm e.
ClisterForm ulaton, M aintenance and U pdate M Eahaniот s
Each CH is rEspons:blE forperiodi:allybrhadcastng it existence utilizing a controlled flooding m echanii . According to this approach, at a certain tim e interval the CH broadcasts a M SolcitateCH message. Thisme^agecontainsitsMAC address (which isal^the group identifier), itsupdated KPI and an auto incram ent number that is used forayc]E prevention. The goal forthis solicitation message is threefold:
(a) Cluster Form ulaton & M aintenance: U pon the receipt of a M SolcitateC H m e^age by a node not registered to a CH , a M RegisterNide2CH response m e^age is send to the CH . The Hatter updates its routing cache and then the node auttom atically becom es a mem ber of the broadcasted cluster. If the node belongs to the cluster controlled by the CH generating the m e^age, the node forwards it to its neighbours. In addition, it stores the m e^age to a local cache in order to avoid ^rving the sam e m e^age again. Otherwise, if the node does not belong to the cluster controlled by the CH generating the m essage, the node belongs to the borderline between two clusters, as shownin Figure 1. In this case, the node does not forward the m e^age to its neighbour but instead forwards it directly to its CH . This proce^ is very critical as it prevents the unnecessary flooding out of the scope of a cluster and allows CHs to be aware of their neighbouring CHs. The routing information, collected to the CHs' caches during the clusters formulation, facilitates the hierarchical routing m echaniOT , as disou^ed in Section 2 3. The cluster form ulation and m aintenance processisaLto shown in Figure 2.
Figure 1. Controlled Flooding M echanism .
Cluster Border
Figure 2. Solicitation M eChanim .
(b) Routing Cache M aintenanœ: Each M SoliciateCH m essage—broadcasted by a CH and flooded within the c]ustel^conlains a lstof the MAC address of the nodes that have already forwarded it since each node that ssives ttnis m essage appends its M AC to this list (only once due to cycle preventive m echanim ). This inform atonalllow s cluster nodes to learn (or update) the shortest path tow ards their-CH and store this inform ation in their local routing cache.
(c) Cluster Formulation Update: The M SoliciateCH message allows nodes to be dynamically distributed am ong the existing clusters according to the CHs KPI In case that a node receives a M SoliciateCH message from a neighbouring CH, it compares the received KPI with the KPI of its аullaneCH . W hen this com parison overcom es a specified threshold, the node unr^i^^ from its cunrent CH ándr^i^^ to the new CH . These tasks are accom pliäied w itththe usage of a M Reg;stelNode2CH and M üriRagistelNodeframCH m essage, accordingly. This approach allow s CHs to extend their lfstim e since the load is re-distributed am ong the m ole pow erful CH s.
Figure 3. Perbdic M echanim forCH election.
The NEURON clustering m echanism allows the autonom ic re-formulation of clusters and enables the adaptation of the clusters' number and formation to the existing network conditions. This m echanism is initiated either because a CH decides to return to norm al operation m ode or because the CH leaves the network due to an unforeseen situation. In the firrstcase, when the KPI of the CH passes below a specified threshold, the node stops to undertake the role of CH and a M SolicitateCHDO W N m essage is flooded within its cluster. N odes update their routing cache and inform neighbouring CHs, provided that routing information for them is available in their routing cache. The re-clustering mechanism is then invoked and the cluster members decide to become a CH with the current probabiltyPclist. If no CH is elected, the nodes pin a neighbouring cluster after receiving a neighbouring M SolicitateCH message. In the second case, if the M SolicitateCH message is not received within a period, the node updates the routing cache and initiates the re-clustering m echanism as previously Figure 3). It should be noted that the CHs rem ove any entries from their routing cache related with neighbouring CHs if no relevant M SolicitateCH message is received within a Specific period.
3 3. Hierarchical Routing
Routing and clustering m echanism s are interrelated inW SN s, bothof them targeting to m inim ise energy consum ption. It is desirable thatpacket forwarding and routing protocol overhead is distributed am ong all the sensor nodes according to their KPI values. This approach preserves scarce sensors recourses and, thus, extends the network lifetim e.
Energy efficiency in N EU RO N is achieved by hierarchical reactive routing. N odes are organized into a hierarchy of clusters based on network proxim ity to the CHs. There is no proactive mechanism to build and m aintain a valid routing table as the network topology continuously evolves. In addition, routing mechanism takes advantage of the routing cache entries generated during the clustering process, as presented in Section 3 2.
Figure 4. Routing in NEURON .
MRouteRequest ~~ AmICH?
Yes I No
Forward2MyCH
The routing algorithm in NEURON presents similarities with the DSR routing protocol [40]. NEURON adopts some mechanisms from DSR torcommunicationamong CHs, while intra-cluster com m unicatbn is designed independently. A RouteRequest m essage is used fordetecting a valid route
to a destinatonnode, in accordance to the DSR protocol, m NEURON , however, the RouteRequest m essage is not flooded but directly forwarded to the CH of the transm itting node. The exactpath to the CH isknown via the M SolicitateCH m essages broadcasts by CHs in rsguJarintervals.
Figure 5. Route RequestM echanism .
W hen a node desires to establish communiation with another node, it initially sends a RouteRequest m essage to its CH Figure 4). The m essage contains the exact path towards the CH (source routing). IE a node receives a RouteRequestm essage, it initially checks whether it operates as a CH . This is necessary because the election of new CHs isa dynamic process and thus new CHs may be present. IE the node is not a CH, then it forwards the m essage to the next hop towards the CH according to the dissam inated path from the initiator node, in case that a node along a path is unreachable, a RouteError m essage is generated and broadcasted within the scope of the cluster that contains the broken link:. If the m essage is delivered to a CH during its path, the CH queries its local cache for the requested route towards the destination.. If the validentTy is found, a RouteResponse m essage is sent to the initiating node. O therwise, the CH forwards the RouteRequestm essage to its known CH s, exactly as routing is implemented in DSR . The CH of the destination node will directly reply to the initatornode with the corr^en^-to-end path Figure 5).
W hen a route to a destination is know n then a M essage^ransfer m essage is in^itia^t^. This m essage contains the source node, the destination node, the route that m ust be follow ed in the W SN in order to reach the destinationanda flag that inform s the destinationnode whether it shouHdrespondw itha confirm ation (acknow ledgem ent). The M essage_Transfer m essage is used for transferring upper layer data, e g , overlay topology form ulation..
3.4. Requiem ente from Sensor N ode Platform s
No special requiements are imposed by NEURON in order to be applied to existing and next generation sensor mots. in NEURON , nodes may be identified according to their 6LoW PAN address while low power 80215.4 radio may be used for communication among them . According to
6LowPAN, IEEE 802.15 4: devices may use either IEEE 64 bit extended addresses or 16 bit addresses tthatare unique within a PersonalArea Network (PAN).
NEURON, as stated earlier, necessitates the storage of routing inform ation in the routing cache of each node. This information regards the routes that are stored in each node during the cluster form ulation and m aintenance process. C onsidering that the m axim um depth (distance in hops) from a CH isapproxim ately five, the size of each routing entry in NEURON varies from 10 to 15 bytes (when nodes are identified by their 16 bit addresses) or from 20 to 45 bytes (when nodes are identified by their IEEE 64 bit extended addresses). Thus, in case of a routing cache with 1,000 units, the m axim um m em ory thatcouLld be necessary is 45 kB (in case that all the routes are m ore than fourhops) while in a realistic scenario 15 kB are sufficient, as also shown in Section 5 2.
Requirements tor energy depend on the type of the sensor mote. However, the behaviour of NEURON regarding energy efficiency is presented in detail in Section 5. Finally, NEURON does not impose special processing requirem ents or constraints forthe operating system used, since NEURON m echanism spertorm sim ple functionalities thatm ay be im plem ented in each operating system .
4. Service Provisioning O ver NEURON
The creation of an overlay topology is beneficial for decentralized and autonom ic service provisioning in W SN s. M ultiple challenges, though, have been identified due to the dynam ic network characteristics in such networks. P2p protocols may efficiently manage the sensors' interconnections (as nodes continuously join or leave from the overlay network) or control the autonom ic delegation of tasks am ong participating nodes [10,42]. A ¿¡dressing scalability and com plexity challenges, though, is stila research issue [49] .NEURON exhibits the necessaryfunctonalityto upper layer protocols in order to establish an overlay network while aim s to address com plexity and scalability issues. It also provides routing inform ation to topology form ulation m echanism s in order to im prove their efficiency
Several topology formulation algorithms have been proposed in the literature [41,46,47]. The T-M AN algorithm was selected for investigating the advantages that NEURON may offer" to the overlay topology formulation mechanism s due to the faster convergence capabilities of T-M AN compared to other alternatives for the creation of an overlay topology [50]. T-M AN allows communications with any node in the overlay network contrary to other gossiping protocols that permit communications with only the one-hop-away neighbours (refer to Section 31). The latter approach increases routing overhead as one node has to identify a proper route prior to attem pt to communicate withanothernode inthe overlay network. 3hT-M AN , each node aim s to identify its successor in the overlay topology based on the know ledge that acquires through the exchange of view s with its neighbours. Scoring functions are applied for the selection of the successor of each node and the results are stored ina buffer. The scoring functionaffects the form ationof the overlay network topology. In our case, a ring topology is formulated since Chord [42] was selected as a p2p protocol for the provision of storage and retrieval functionality over the created overlay network:. Chord pre-assum es that nodes are ordered in a ring and are aw are of their successor and predecessor in the overlay ring topology. However, any other topology may be also aeated based on the selected p2p protocol.
Figure 6. Topology FormulationAlgorithm .
Table 2. T-M AN pseudo code.
do forever( Node To Sentp ^ selectCloserNode() buffer ^ merge(view,(myNodeDescriptor}) send buffer to Node To Sentp receive bufferp from Node To Sentp buffer ^ merge(bufferp ,view) view ^ Reevaluate(buffer) } do forever{ receive bufferq from Senderq buffer ^ merge(view,(myNodeDescriptor}) send buffer to Senderq buffer ^ merge(bufferq ,view) view ^ Reevaluate (buffer) }
active thread passive thread
T-M AN applies a gossping technique [50] in order to identify the relative position in the overlay (Table 2). Each node m aintains a view w iththe nodes that are—up to a specific tim e—known and scored. Each node periodically communicates with the "closest" node and exchange views with it Figure 6ab>) - A ftr this m utual exchange, nodes re-evaluate their view s Figure 6cd). This iterative procsdurs leads to extrem ely fast convergence, ie, the state in which each node know s its successor. A ny further exchange of m essages betw een the nodes does not dm prove the accuracy of their view s Figure 6e).
The T-M AN algorithm is adapted in order to exploit inform ation available through the NEURON m echanisrn s and thus m inim ize the m essages that are necessary for convergence of the protocol in a W SN . Two improvem ents have been integrated to the T-M AN algorithm . Firstly, each node, before sending its buffer to a requestor, updates the buffer with nodes that score better than the existing ones by consulting its routing cache. Secondly, m ultple m essages are sentto any node in the buffer instead of sending one m essage to the first node of the buffer (ie. the node that scores better. This facilitates the fast dissem ination of inform ation regarding the network topology which is critical in dynam ic networks. These t o adaptations reduce the total am ount of m essages required for overlay topology form ulation., as show n in Figure 7.
Figure 7. Total m essages foroverlay topology form uHaton.
Afterthe overlay network is establllii^Lecl, participating nodes are able to store and retrieve data using typical p2p protocols. Every node that aim s to access the p2p network storage (ieto store a key/value pair or query a value based on a key) may use a Distributed Hash Table (DHT) [51] that operates on-top of the overlay topology. Inour sim ultinexperim ents, Chord [42] w as selected for integrating DHT functionality. Chordis an efTrienLtdistributed lookup system based onLCbnLsistenthashing. Its only operationL is to map a key to a responsible node. Chord scales well with a number of nodes and, thus, it can be applicable to large networks. It continues to functonL correctly even if the system undergoes m ajor changes or if the routing inform ation is partially correct [42].
Figure 8. Autonom ic Provision of Services.
Advanced services may be builtusing two API functions putkey value) and get(key), which interact directly with the DHT protocol Figure 8). For instance, a distributed storage system handling envionm ental m onitoring data m ay be built by using com m on hashing functions and predefined keys [42]. Envionm ental data is consequently available for retrieval and further processing by all networknodes. The prvidedservices could be d^cen^tralizedais data and n^c^ssaryfunctionalityis allocated in multiple nodes at the overlay network:. If necessary, som e critical functions may be delegated to more than one nodes for improving reliability. In case of network changes or node failures, roles may be re-assigned autonomously and performance guarantees may be assured for the services provision.
5. E xperim ental E valuation
In this section the perform ance of NEURON for a wide set of topologies is evaluated. NEURON has been developed in the Peersim simulator [52]. A visualisation module is also developed as a Peersim extension that provides a view of the clusters with their CH s that are form ulated in the W SN at each cycle periodof the sim ulation. Inorder to simulate the lim ited resources of the participant nodes of the W SN ,a custom dynam ic m odel is incorporated that im poses penalties according to the nodes operations.
In the sim uHations m ultple nodes are sim ultaneously activated w ithout any preconfigured state information.. Each simulationlasts 2,000 cycles, while every node is -initialized with 100,000 battery units and 1,000 routing cache m em ory units. B attery and m em orypenalties are defnedfor serving each m essage in the network:. Each entry in the routing cache occupies one m em ory unit, each packet transm ission or reception drains the available battery by three units, while each packet processing action (e g . protocol encapsulation) that is accom plished by a node drains the battery by one unit. The periodic broadcasting of M SolicitateCH m essages is set to five cycles, the KPI threshold refer to Section 3 2) for transition to a new CH is set to two, and the threshold where a CH sw itches to norm al m ode (referto Section 3 2) is set to 50% . A ll the nodes are considered with equal battery and m em ory capabilities at their initial deploym ent The number of nodes varies from 50 to 12,800 while the density varies from three to 36.
The perform ance of each mechanism is assessed using multiple criteria, such as messages exchanged for the operation in steady state, convergence capability, precision in the estim ation of param eters, behaviour of the probabilistic techniques, energy efficiency and quality of distribution of CHs in diverse network sizes and densities. Sim ulation results that are related with the creation of the overlay network and NEURON's suitability for deploym ent of advanced services in the W SN are also presented. E ach sim ulation is executed five tim es and average values are considered in our analysis.
51. Evaluation of the Autonom ic Estimation Algorithm
The Autonom ic Estimation Algorithm , aim ing to estim ate two network param eters, is activated right after the sensor nodes becom e operational. In each node, the algorithm converges if the estim attd parameters' values between two consecutive cycles is less than 5% .
Figure 9 a) show s the num ber of m essages that are exchanged until the algorithm converges to the estim ation of the density and the size param eterfor-various sizes and densities. Figure 9 (b) show s how these m essages are distributed betw een nodes. There is a linear relationship betw een the total num ber of m essages exchanged and the network size. It derives that the algorithm convergences w ihout im posing significant overhead since the average num ber of m essages per node rem ains sm all and stable even for large-scale networks. The autonom ic estim ation process m ay be repeated periodically in predefined num ber of cycles, related with the dynam icity that is present in the W SN .
Figure 9. a) M essages tor NetworkDensityand Size Estimation and (b) M essages tor Network Density and Size Estimation pernode.
Figure 10. Cycles forconvergence of the estim attd parram eterrs.
Network Size
Figure 10 presents the cycles that are necessary for the algorithm to converge. Stable behaviour is achieved for each network density independently from the network size. The algorithm converges in less than 20 cycles in sparse networks and in less than 10 cycles in dense networks, independently of the network size. In dense networks, convergence is taster since more m essages are exchanged at each cycle.
The Autonom ic Estimation Algorithm achieves adequate precision for the estim ation of the network param eters, as presented in Figure 11 (a) for the density estim ation and Figure 11 (b) for the size estim ation. For the density estim ation, the devdatonL from the real values is less than 9% in all cases
and rem ains approxim ately constant for a given network density. For the size estim aton, the deviation is Hess than 20% for sparse networks and less than 10% in dense networks. In both cases, higher precisionis noticed indense networks since averaging is perform edbetween m ultiple neighbours in each cycle.
Figure 11. Deviation in the estimation of the (a) density and (b) size of the W SN .
5 2. Clustering and Routing M echanism Evaluation
A visualisation m odule is developed for the dynam ic SHustration of the clustering process and the distribution of the CHs within the WSN. It is noticed that the clusters' distribution improves (qualitative m etric) over tim e since the probabilistic techniques used tend to hom ogenize the size and form of the created clusters and thus distribute the clustering overhead among the elected CHs. Two indicative saeenshots are presented in Figure 12 where clusters are distinguished.
Figure 12. CHusterVisualisationL.
O Cluster 1 O Clister2 O CHs
O Cluster O CHustBr2 O Cluster3 O CHs
In Figure 13, the com parison of the num ber of CH s that are elected in the sm ulation envionm ent with the theoretical ones according to the Pciust refer to Equation 1) is shown. As it is expected, the theoretical and sm ulation results are closely related.
Figure 13. Theoretical and Practical num berof elected CHs.
Network Size
An im portant characteristic for the optim isationof the clustering process is the adaptationof the clusters' size according to the changes in the network topology. In cases of more dense networks, it is desirable the cr^tionof larger insize) clusters since nodes are close (innum ber of hops) to each other:. The existence of less CH s w ith sm all distances from their m em bers im proves the energy efficiency of the W SN refera!^ to Section 53). In Figures 14(a)-(c), the average size of the clusters that are created is presented for fixed and variable probability (r^^to Equation 1). In the latter case, the trend is the creation of larger in size clusters in dense networks in opposition to the firstcase where the cluster size remains stable. Self-optimisation of the clustering process is ther^re achieved. However, m ore optim al equation for the Pclust probability may be selected, in case that sm aller clusters are desirable in dense networks. Furthermore, great variation is present in small-scale networks Figure 14 (c)) due to the im pact that has the probability in the cluster size, as the num berof the elected CH s significantly affects the average cluster size. This variation is decreased as the period that NEURON is applied in the W SN increases since probabilistic techniques follow an optinalbehaviour.
Figure 14. ClisterSize for a) Pclist = 1% , b>) Pdust = 4% and (c) variable Pclist-.
(a) (b)
Network Si,-:. Network Size
Figure 14. Cont.
Network Size
The average number of route entries that are stored on each node's routing cache after the cluster form ulatbnprocess is show ninFigure 15. This num ber is critical since sensor nodes m ay present m em ory constraints. Pclust is variable according to Equation1. ]tis shownthat the num ber of route entries increases slightly as the size and the density of the network increases. H ow ever, this num ber rem ains bounded, even for large networks. In sparse networks, an average routing cache has 50 entries, while in dense networks an average routing cache with 150 entries is needed. Since the m axim um size of a routing entry is 45 bytes (refer to Section 3.4), 15kB of routing cache size is adequate for all the sensor nodes.
Figure 15. Average entries in each node's routing cache.
0i_i_i_i_i_i_i_i_i_i_i_i_i_i_
0 2000 4000 6000 8000 10000 12000 14000
H additionto the num ber of route entries, a qualitative m etc is the percentage of the total route entries that exists in the CH's routing caches since these entries are used from the routing functionality in NEURON . This m etric is depicted inFigures 16 a)- (c) tor fixedand variable probability. As the num ber of CHs increases in the network, the percentage of the total routing entries that exist in their routing caches also increases. This is shown in Figures 16 (a,b), where the selection of larger value for the stable probability results to higher percentages of routing cache entries in the CH s. W hen this percentage is sm aller, intra-cluster comm unication is facilitated since the nodes that are not CH s have
avaiab]e routes for other nodes within the cluster and do not need tto com m unicate with their CH for establishing a route tow ards them . This percentage is sm aller for dense networks due to the greater overlapping am ong the cluster zones and the existence of m ultple paths tow ard a CH . N odes intthe overlapping regions store routing entries towards m ore than one CH . The percentage also increases as the size of the network increases while when a variable probability is used Figure 16 (c)), the percentage is sm all for dense networks and large for sparse networks.
Figure 16. Percentage of total routes in the CHs routing caches for (a) Pclist-(b) Pclust = 4% and (c) variable Pclust.
In Figure 17 (a), the total number of routing m essages that are exchanged until the clustering formulation iscompleted ispr^^t^, while in Figure 17(b), the sam enumberpernode in the W SN is shown. M ore routing m essages are exchanged indense networks, due to the nature of the controlled flooding mechanism that has been adopted from the M SolcitateCH message. Although cluster form ulation m essages are confined and cycl^-pr^en^t^ as discussed earlier, the existence of multiple connections for each node creates an analogous routing overhead that is avoided in sparse networks. Furthermore, inFigure 17 (b) itis shown that the clustering formulation mechanism is scalable since the num berof m essages per node for different densities rem aiin either stable or slightly increases as the network size increases.
Figure 17. (a) Total number of messages exchanged for cluster formulation and (b) total num berof m essages exchanged for cluster form ulaton pernode.
Network Size Network St®
Upon com pletion of the clustering process, routing functbnaliiyexploiis the available inform aton in the routing caches of the se^sornodes. Ih Figure 18 (a) and 18 (b) the total num berof RouteRequest. and RouteRegponse m essages that are exchanged in order a node to identify a valid route, are depicted. The num ber of the generated RouteRequest andRouteResponse m essages is radicallyredu<ced as the network density increases since more routing information is already available in the nodes. Thus, NEURON's scalability is addressed as the routing overhead (in number of messages) is considered low .
Figure 18. &) Route Requestand (b) Route RegoonseM essageCostt
Network Density ~ Network Density
53. Energy Efficiency in NEURON
Simulations are perform ed inorder to assess the energy efficiency of NEURON m echanism s and their impact to the network lifetim e. Network lifetim e refers to the time period where all the nodes of the network (or a very high percentage of them ) are «operational. The network size is set to 1,000 while the initial energy of each node is set to 100,000 units. Sim ulatons are term inated when the existing W SN is golit into two orm ore isolated groups as nodes leave the W SN when their battery is exhausted.
Figure 19. a) Alive Nodes in the W SN, (b) Residual Energy in the W SN aid (c) Consum ed Energy in the W SN .
In Figure 19 (a), the number of nodes that are alive while the number of cycles increases is shown for fixed aid variable probability aid different densities. The threshold where a CH sw itches to norm al mode—when the KPIof the CH goes below it^isset tb 50% . It isnoticed that the network tends tb extBndits networklifBtine as the num ber of alive nodes reduces steeply aftera certain. num ber of cycles. Thismeans that the available power of the sensor nodes is almost the same aid thus they run out of pow er in a few cycles. Furtherm ore, it is show n that the network liffetim e is Longer in the case of applying the -variable probability aid in case of m ore sparse networks. This is reasonable since few er messages have tb be exchanged inspare networks for cluster formulation aid maintenance. In Figure 19 b), the residual energy in the network is shown. The threshold where a CH switches to normal mode is set tb 50% aid 75% while the density is set tb 15 aid 30, respectively. The residual energy is higher in case of sparse networks aid in the case where the threshold is set to 50% . A high threshold reduces the rotation in the CHs in the W SN aid causes high energy consumption in each CH , causing them tb run out of energy earlier than the other nodes. In this case, therefore, energy consum ptbn is not hom ogeneously distrbutBd am ong the sensornodes. In Figure 19 (c), the consum ed energy is depicted in case of the fixed aid the variable probability, while the density is set tb 15 aid the threshold where a CH transits tb norm al m ode is œtto 50% . It is clear that the application of the
autonom ic m echanism in the election of clusters is m ore energy efficient com pared to the application of a stable probability.
5.4. Topology Formulaton M echanim Evaluation
As described in S ecton 4, the NEURON protocol facilitates the creation of an overlay topology and consequentythe deploym ent and provision of autonom ic services over it. In order to show NEURON's suitability for this purpose, w e com pare the m essages that are generated for the overlay topology form ulaton using DSR [40], ie., another reactive routing protocol. In Figure 20, it is shown that the logarithm ic behaviour of routing cost inD SR im poses extrem e overhead to the network in com parson to NEURON . Furtherm ore, this overhead is m uch greater in dense networks.
Figure 20. Routing M essage Cost forTopology Form ulaton (logarithm ic scale).
6. Conclusions
NEURON, an innovative protocol for autonom ic clustering and routing in wireless sensor networks, is presented in this paper:. Self-configuration and self-optm isation properties are supported by the proposed set of m echanism s. The use of probabilistic techniques com bined w ith decentralized approaches and cooperation am ong nodes for dissem inaton of useful inform atton provide reliability, robustness, energy efficiency and scalability in NEURON's mechanisms. In addition to the deploym ent of autonom ic functionalities within the W SN, NEURON facilitates the creation of overlay topologies over the W SN. w ithout im posing significant overhead. O verlay networks m ay be proven extrem ely useful for the developm ent of advanced services in the sensor networking world, taking into consideration the vision fora W orld-W ide SensorW eb.
The behaviour of the protocol is evaluated according to a wide setof sam ulatons. It could be argued that NEURON behaves well for autonom ic setup» and m aintenance of clusters while inform atton from network-based estm atton techniques m ay be used for optm isation purposes. Routing inform atton collected during the clustering process proves to be valuable since the num berof m essages that have to be exchanged for communication among tile sensor nodes is reduced. NEURON does not impose severe requirem ents for m em oryusage inthe sensor nodes andachieves significant extensdontothe network liffetm e through the rotation of CHs in the W SN field.
In our future work, each of the fundamental mechanics of the NEURON protocol will be com pared w ith other existing protocol's m echanism s. The behaviour of the described probabilistic techniques will be studied in detail, and possible optim isationsmay be proposed for diverse network topologies. Furtherm ore, the perform ance of som e indicative services provided over the overlay network will be exam ined. Finally, the efficiency of m echanism s that create topology-aware overlay networks overNEURON will be studied.
A cknow ledgem ents
This publication is based on work partially performed within the framework of the European Commission ICT/FP7 EFIPSANS project (www efipsansorg).
References
1. Brasche, G . Trends and Challenges of W ireless Sensor Networks, m Proceedings of the Eighth International W orkshop on Applications and Services in W ireless Networks, K asseH, G erm any, October26-30, 2008; p. 7.
2. Rajeev, S; Ananda, A.; Choon, CM .; Tsang, OW . Mobile, Wireless;, and Sensor Networks—1Technology,Applications, and Future Directions; John W iley & Sons: Hoboken, NJ, USA, 2006.
3. Hui, JW ; Culler, D E . Extending IP to Low -Power, W ireless Personal Area Networks. IEEE InternetC om put. 2008, 12, 37-45.
4. Ali, M ; Langendoen, ZK.A Case forPeer-to-PeerNetworkOverlays in Sensor Networks. In Proceedings of the W orkshop on W ireless Sensor Network, Delft, The Netherlands, January 2007.
5. Dressier, F. A Study of Self-Organization Mechanismsin Ad Hoc and Sensor Networks. Com put. Commun. 2008, 31, 3018-3029.
6. M ills!, KL. A Brief Survey of Sef-Organization in W ireless Sensor Networks. W ireless Com m un. M ob. Computt 2007, 7, 823-834.
7. Diaz, G. Autonomic Networks and Management of Dynamic Services Deployment^A Study of Approaches by Using Overlay Networks. In Proceedings of the IEEE Global1Telecom m unications Conference, New Orleans, LA, USA, December2008.
8. Chatterjea, S; Nieberg, T; M eratnia, N ; Havinga, P. A Distributed and S elf-Organizing S cheduling Algorithm for Energy-Efficient Data A ggregation in W ireless S ensor Networks. ACM Trans. Sen. Netw . 2008, 4, 1-41.
9. Jin, S ; Li; K . LBCS: A Load Balanced Clustering Scheme in W ireless Sensor Networks. In Proceedings of the Third international Conference on the Multimedia and Ubiquitous Engineering, Qingdao, Shandong, China, June 2009; pp. 221-225.
10. Lua, E K ; Crowcroft, J; Pias, M ; Sharma, R ; Lim, S. A Survey and Comparison of Peer-to-Peer Overlay Network Schem es. IEEE Commun. Surv. Tutorials 2005, 7, 72-93.
11. Biskupski, B .; Dowling, J.; Sachau, J. Propertes and M echanLism sof SeHf-Organüng M ANET and P2P Systems. ACM Trans. Auton. Adapt. Syst. 2007, 2, 1-34.
12. Hora, D N .; M aaedo, D F.; Oliveira, LB .; Sigueira, IG .; Loureio, A A .; Nogueira, JM .; Pujolle, G . Enhancing Peer-to-Peer Content D iscovery Techn±[ues O verM obile Ad Hoc Netw . Com put. Commun. 2009, 32, 1445-1459.
13. Hu, Y C .;Das SM .; Pucha, H . Peer-to-Peer Overlay Abstractons in M anets. In Theoretical and Algorithm ic Aspects of Sensor, Ad Hoc W ireless andPeer-to-Peer Networks; CRC Press: Boca Raton, FL, USA, 2005.
14. Hu, Y .; Li, W .; Kang, Z . Study on Energy Effcient H ierarchLicalRouting Protocols of W ieless Sensor Network:. In Proceedings of the In^ternational Conference on In^tbrmaton Engineering, Tadyuan, Shanxi, China, JuJy 2009; pp. 325-328.
15. Chamam , A .; Pierre, S.A D istributed EneTgy-Efficient Cluster Formation Prottocolibr W ireless Sensor Networks. In Proceedings of the 6th IEEE Conference on Consumer Communications and Networking Conference,Las Vegas, NV , USA , January 2009.
16. Chaudhry, SA.; Jung, W D .; Akbar, A .; Kim, KH. Proxy-Based Service Discovery and Network Sefecton in 6LoW PAN . Lect Notes Com put. Sci. 2006, 4208, 525-534.
17. M arsh, D .; Tynan, R .; O 'Kane, D .; O 'Hare, G M P. Autonom ic W ireless Sensor Networks. Eng. Appl. Artf. InteH. 2004, 17, 741-748.
18. Rabiner, W .; Chandrakasan, A .; Balakrishnan., H . Energy^-Effic^i^tCommun^ica^tibnProtocol ibr W ireless M icrosen^sor Networks. In Proceedings of the IEEE International Conference on System Sciences,M aui, HI,USA ,January 4-7,2000.
19. Kochhal, M .; Schw iebert L.; Gupta, S . Role-basedH ierarchLicalSelf O rganizationfor W ireless Ad hoc Sensor Networks. In Prcceedings of the 2nd ACM Iltematonal Conference on W ireless Sensor Netw orks and Applications, ACM Press!: San Diego, CA , USA, 2003; pp. 98-107.
20. Estrin, D .; G irod, L .; Potte, G .; Srivastava, M . In.strum enting the W orld w ithW irreleess Sensor Networks. In Proceedings of the International Conference on Acoustics, Speech, and Signal Prooessdhg, Salt Lake City,UT,USA, M ay 7-11,2001; pp. 2033-2036.
21. Heinzelm an, W R .; Chandrakasan, A .; BalakrislnnanL, H. An ApplcatinL-Specific Prbtxbl Archiecture for W ireless M icnbsenscr Networks. IEEE Trans. W ireless Commun. 2002, 1, 660-670.
22. Huang, G .; Li, X .; He, J. Emergy-Eff:ci:iehcy Amalysis of Cluster-Based Routing ProtbCbls in W ireless Sensor Networks. In Proceedings of the IEEE Aerrospace Conference, Big Sky, M T, USA ,M arrch 4-11, 2006.
23. Fonseca, R .; Ratnasam y, S .; Zhao, J.; Culler, D .; Shenker, S .; Stoicai, I. Beacon Vector Routing: Scalable Point-to-Point Routing in W ireeless SernsornLets. In Proceedings of the 2nd Symposium on Networked Systems Design and Implementation, Boston, M A , USA, M ay 2005.
24. Mao, Y .; W ang, F.; Q iu, L .; Lam , SS.; Sm ith, JM . Sm all State and Sm all Stretch Routing Prctbcbl for Larrge W irreleess Sensor Networks. In Proceedings of the 2nd Symposium on Networked Systems Design and Implementation, Cambridge, M A , USA , April2007.
25. Iwanicki, K .; Van Steeen, M . On Hierarchica! Routing in W ireless Sensor Networks. In Proceedings of the Untematiornal Conference on :Ih.:brm aton Processing in Sensor Networks, San Francisco, CA , USA, April2009; pp. 133-144.
26. Ibriq, J.; M ahgoub, I. Clust^BasedRouting in W irelessSensor Networks: Ussuesand Challenges. In PrccsEdings of the 2004 Symposium on Performance Evaluation of Computer TelECommunicatbn Systems, San Jose, CA , USA, JUJy 2004; pp. 759-766.
27. Jiangs, C .; Yuan, D .; Zhao, Y . Towards Clustering Algorithm s in W ireless Sensor Networks—A Survey. In Proc^^ings of the IEEE W ireless Communications and Networking Co^f^^c^, Budapest. Hungary, April 2009.
28. Abbasi, A .; Younis, M . A Survey on Clustering Algorithms for W ireless Sensor Networks. Com put. Commun. 2007, 30, 2826-2841.
29. Al-Karaki, J; Kamal A . Routing Techniques in W ireless. Sensor Networks: A Survey. IEEE W ireless Commun. 2004, 11, 6-28.
30. Younis, M ; Mun^ P; Gupta G ; Ellshaikawy, SM . On Efficient Clustering of W ireless Sensor Networks. mProceedings of the 2nd IEEE W orkshop on D ependabilty andSecurty inSensor Networks and Systems, Pisa, Italy, April 2006; pp. 78-91.
31. Younis, O ; Fahmy, S . HEED : A Hybrid, Energy-Efficient, Distributed Clustering Approach for Ad Hoc S e^sor Networks. IEEE Trans. M ob. Com put. 2004, 3, 366-379.
32. W ang, Y ; Zhao, Q ; Zheng, D . Energy-DrivenAdaptive Clustering Data CollectionL Protocol in W ireless Sensor Networks. In Proceedings of the International Conference on Intelligent-M echatrQnics and Automation, Chengdu, Sichuan, China, August26-31, 2004; pp. 599-604.
33. W u, Y ; Chen, Z ; Jing, Q ; W ang, Y . LENO : LEastRotetionLN ear-OptimalClusterH ead Rotation Strategy in W ireless Sensor Networks, m Proceedings of the 21st ]hternational Conference on Advanced Inform ation Networking and Applications, Niagara Fals, ON, Canada, M ay 2007; pp.195-201.
34. Lindsey, S; Raghavendra, CS. PEGASIS: Power Efficient Gathering in Sensor mformationL System . In Procsedings of the IEEE ConLferEncs on Aerospace, Big Sky, M T,USA, M arch 2002; pp.1125-1130.
35. Ye, M ; Li; C ; Chen, G ; W u, J. EECS: An Energy EfficientClustering Scheme in W irel^ Sensor Networks, m Proc^^ings of the 24thIEEE Intemational Perform ance, Computing, and Comm unicationLs C onLferencs, Phoenix, AZ, USA, April2005; pp. 535-540.
36. Lee, H ; Lee, K ; Lee, Y . A Self-Organized Head SeIecebh for Hierarchical Routing in W ireless Sensor Networks. In Proc^^ings of the 3rd IntErnational W orkshop on Self-Organizing Systems, Vienna, Austria, December2008.
37. W ang, Y H ; Tsai, CH; Mao, H J. HM RP: H irarchy-Based M ultipath Routing Protocol for W ireless S ensor Networks. In Em bedded and Ubiquitous C om puting; SprjigerBerlinHeTdelherg: Taipei, Taiwan, 2007; pp. 452-459.
38. Simon Carbajo, R ; Huggard, M ; M cGolldr±k, C. An End-to-End Routing Protocol for PeEr-to-PEsr Communicationi in W ireless Sensor Networks, m Proceedings of the 6th W orkshop on M iddleware forNetwork Eccentric and M obie Applications, Glasgow , Scotland, April 2008.
39. Jayakumar, G ; Gopinath, G . Ad Hoc M obie W ireless Networks Routing Protocols—A Review . J". Com put. Sci. 2007, 3, 574-582.
40. Johnson, D ; M altz, D ; Broch, J. Ad Hoc Networking; Addison-W esley: UpperSaddle River, NJ, USA, 2001, pp. 139-172.
41. Bicocchi N .; M am ei, M .; Zambonelli, F. Self-O rganizing Overlay Knowledge Networks for Future Pervasive Computing Applications; Am erican Scientific Publishers: Valencia, CA , USA , 2007.
42. Stoica, I.; Morris», R.; L ien-Nowell, D .; Karger, DR.; Kaashoek, M F.; Dabek, F.; Baiakrishnan; h . Chord: A Scalable Peer-to-Peer Lookup Protocol for Internet Applications. IEEE/ACM Trans. Networking 2003, 11, 17-32.
43. Rowstron, A .; Druschel, P. Pastry: Scalable, Distributed Object Location and Routing for Large-Scale Peer-to-Peer Systems. In Proceedings of the 18th IFIP/ACM International Conference on Distributed System s Platform s, Heidelberg, Germ any, Novem ber2001.
44. Jung, CT.- Lee, S .; Kim, N .; Y oon, H . Efficient Service Discovery M echanism for W ireless Sensor Networks. Com put. Commun. 2008, 31, 3292-3298.
45. M uneeb, A .; Zartash, A . CSN : A Network Protocol for Serving Dynam ic Queries in Large-Scale W ireless Sensor Networks. In Proceedings of the Annual Conference on Communication Networksand Services Research, Fredericton,NB , Canada, M ay 2004; pp. 165-174.
46. Friedman, R .; Gavidia, D . Gossiping on M ANETs: The Beauty and the Beast. ACM SIGOPS Oper. Syst. Rev. 2007, 41, 67-74.
47. Ding, W .; Iyengar, SS. Bootstrapping Chord over M ANETs—All Roads Lead to Rome. In Proceedings of the IEEE W ireless Communications and Networking Conference, Hong Kong, China, M arch 2007.
48. Jeksity, M .; M ontresor, A.; Babaoglu, O. Gossip-Based Aggregation in Large Dynamic Networks. ACM Trans. Comput. Syst. 2005, 23, 219-252.
49. Sam sudin, A .T.; Suryana Herman, N .; Unger, H .; Awang, M TK. The Searching Scalability of Peer-to-Peer System . In Proceedings of the 10th International Conference on Advanced Communication Technology, Gangwon-Do, Korea, February 17-20, 2008; pp. 1891-1896.
50. Jeksity, M .; M ontresor, A.; Babaoglu, O . T-M an: Gossip-Based Fast Overlay Topology Construction. Com put. Networks 2009, 53, 2321-2339.
51. Kang, E .; Kim, M J.; Lee, E .; Kim, U. DHT-Based Mobile Service Discovery Protocol forM obile Ad Hoc Networks, In P roceedings of the 4 th International C onference on Intelligent- c om puting, Shanghai, China, Septem ber2008; pp: 610-619.
52. Jekcity, M .; M ontresor, A .; Jesd, G .; Voulgaris, S. The Peerem simulator. Available online: http //oeersm fnet (accessed on M ay 2010).
© 2010 by the authors; license M DPI, Basel, Switzerland. This article is an Open Access article
distributed under the terms and conditions of the Creative Commons Attribution license
(http //creativecom m onsorg/icenses/oy/B .0/).
Copyright of Sensors (14248220) is the property of MDPI Publishing and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.
