PAPER. 1. Introduction

Transcription

1 IEICE TRANS. COMMUN., VOL. E9x-B, No.8 AUGUST 2010 PAPER Integrating Overlay Protocols for Proviing Autonomic Services in Mobile A-hoc Networks Panagiotis Gouvas, IEICE Stuent member, Anastasios Zafeiropoulos,, IEICE Stuent member, Athanassios Liakopoulos, Gregoris Mentzas an Nikolas Mitrou 1 SUMMARY Next generation network characteristics increase the complexity in the esign an provision of avance services, making inappropriate the selection of traitional approaches. Future networks are becoming larger in scale, more ynamic an more heterogeneous. In orer to cope with these requirements, services are expecte to aapt to environmental conitions an require minimum human intervention. In this paper a new moel for proviing autonomous an ecentralize services is propose, especially focusing on mobile a hoc networks (MANETs). Using a newly propose four-layere approach, service evelopment may be realize inepenently from the unerlying physical network. In a reference implementation, it is emonstrate that it is possible to set up an overlay network that hies any network changes from the service layer. Multiple mechanisms have been aapte in orer to efficiently in terms of message exchanges an convergence time operate over an a hoc environment. Finally, it is emonstrate that a specific service coul operate over a ynamic network with multiple failures. key wors: Autonomic services, MANETs, aapte T-MAN, DSR, p2p, overlay networks. 1. Introuction The tren towars the Future Internet is the transition to networks that provie highly istribute, pervasive an communication-intensive services. These services pose new requirements on the unerlying network infrastructure [1-2] that erive from the nee to manage ynamic, heterogeneous an complex networks an to enable seamless transition from current to future networks. In orer to cope with these requirements, such services will be expecte to (i) be aware of the context an the environment in which they operate, (ii) selfconfigure an self-aapt accoring to the network conitions that they sense an (iii) require minimum feeback from the en-user an no explicit human intervention. The incorporation of autonomic principles for the transition from a service agnostic Internet to a servicean self-aware Internet is necessary [3]. Autonomicity is require in orer to accomplish management of ynamic, heterogeneous an complex networks, where each network entity nees to be able to take network Manuscript receive December 03, Manuscript revise March 1x, The author is with National Technical University of Athens, Heroon Polytexneiou, 15773, Zografou, Greece. The author is with Greek Research an Technology Network, Av. Mesogion 56, 11527, Athens, Greece. optimization ecisions. The monitoring components of an autonomic network shoul be continuously aapte, in a flexible manner, to an ever changing network infrastructure without ignoring the goals an the constraints that the aministrator has set for the network as a whole. These challenges can be aresse through the creation an maintenance of overlay networks [5]. Overlay networks create a well-efine virtual topology above the unerlying network infrastructure, where autonomic functionalities can be implemente an avance services may be provie. However, new methoologies for implementing an operating overlays are neee. In particular new mechanisms are require that permit overlays to structure their topology, efine their routing scheme, an manage their resources inepenently [4-5]. The transition to overlay networking an the incorporation of autonomic characteristics in next generation networks is most crucial in a-hoc an mesh networks since they present specific challenges, especially ue to the following characteristics: (i) ynamic topology (links are establishe / torn own frequently), (ii) unreliable operation (noes may be isconnecte at any time), (iii) complexity (lack of centralize control or network hierarchy) an (iv) loose control of noes (roles in the overlay network are automatically assigne). Peer-to-peer (p2p) networks constitute the best caniate for aressing most of these challenges (unreliable operation, complexity, loose control of noes) [23]. In p2p networks a ynamic assignment of tasks among peers is realize an the provie services are base on their irect cooperation. Peer noes share the available resources, provie reunancy ue to istribution an replication of ata an thus impose minimal requirements to the infrastructure. A-hoc eployment increases the self-organization level of the network an permits its operation also in an unstable environment with loosely connecte noes. Furthermore, espite the apparent chaos of perioic ranom changes to the membership of the network, p2p networks provie provable guarantees about performance. However, the aoption of current p2p techniques necessitates the existence of a fixe network topology [16, 10]. With the term fixe, we refer to wire an wireless networks that have formulate their physical Copyright 2010 The Institute of Electronics, Information an Communication Engineers

2 2 topology an o not present any topology changes. Over a formulate physical topology, specific p2p functions can be applie an a wie range of services can be eploye. This is not the case in MANETs. Each evice in a MANET is free to move inepenently in any irection, an thus frequently change its connectivity to other evices. In this paper an approach is propose for autonomic provision of services in MANETs. Focus is given on issues such as topology formulation, aaptation to topology changes, scalability an stability. Gossiping an exhaustive tree-base techniques for topology formulation are compare an specific aaptations to the routing protocols are propose. The escribe approach is generic enough an can be use for the esign an implementation of autonomic p2p services, from network monitoring an management to chatting an social networking applications. The paper is organize as follows; section two briefly presents the current work in the fiel of autonomic services in MANETs. Section three escribes the basic components an the esign principles of our approach. Section four etails specific implementation issues while section five escribes the evaluation scenario an the simulation results. Finally, section six conclues the paper with a short summary of our work an a iscussion of open issues an future work. 2. Relate Work 2.1 Transition to autonomic service-oriente networks The management of a ynamic, heterogeneous an complex network may become more robust an stable, provie that each network noe is able to take local or network optimization ecisions without strict centralize control [3]. Autonomic capabilities have to be integrate to network monitoring algorithms in orer to overcome the inefficiencies of traitional network management systems usually base on centralize architectures. In an autonomic environment, the network noes themselves are able to etect, iagnose an repair failures, an aapt their behavior accoring to generic network policies [6]. Several approaches have been propose for autonomic network architectures, such as 4D [7], CONMan [8], a Knowlege Plane for the Internet [9] an FOCALE [6]. Our approach, as presente in the following sections, aheres principles from the GANA architecture [11]. In GANA, autonomicity is realize through the esign of control loops operating within network noes/evices an the network as a whole. In our approach, self-* functionalities (e.g. self-organization, self-optimisation, self-healing) are esigne base on interactions among the existing autonomic entities. Interactions among noes in the same networking neighborhoo have been efine while no centralize entities or human intervention is necessary for the provision of services. However, the explicit mapping of our approach to the GANA architecture is out of the scope of this paper. Current efforts inicate that the esign of systems that combine the concepts of self-organization an ecentralization can be realize through the creation of overlay networks over the unerlying network infrastructure [12]. Overlay networks are a very successful paraigm that enables both the construction an the evolution to the future service oriente networks. In [4] it is state that in orer to aress Future Internet challenges new methoologies for implementing an operating overlays are neee with possible re-layering of the existing IP protocol layering. In [3] a new autonomic management architectural moel that facilitates the transition from a service agnostic Internet to service aware network by creating a self-managing virtual resource overlay is escribe. Finally, in [13] an approach is presente for exploiting the autonomic functionalities of peer-to-peer base overlay technology to form an autonomic service control for next generation networks. 2.2 Use of p2p protocols an DHTs in MANETs The optimal way to manage resources automatically in MANETs is through the aoption of techniques that are applie in p2p protocols. This is ue to the reason that p2p an MANETs present many similarities that can be exploite, such as ecentralization, self-organization, an ynamically changing set of noes [ ]. Furthermore, characteristics, such as robustness an reliability, provie by the elegation of tasks to several noes, make p2p techniques a perfect fit for content issemination in MANETs. P2p protocols rely on ecentralize structures, such as Distribute Hash Tables (DHTs). A DHT is a structure that is collaboratively built by all participating noes of a network an provies a lookup service for resources that are publishe by these noes. Resources are utilize transparently an ata is store/retrieve seamlessly in the network. Several existing approaches try to improve service iscovery performance an efficient issemination of content by eploying the concept of DHT in MANETs [17-18]. Limite work has been one for the overlay topology creation an maintenance in MANETs [19-20]. Although many reference implementations exist for p2p protocols -such as Chor [21], Pastry [22], an many others [23]-, these (protocols) operate uner the assumption that a fixe network topology is alreay establishe, on top of which the overlay network is create an maintaine. P2p protocols allow a user to store an retrieve key-value pairs in the network seamlessly without knowing where the actual ata is store. The task of overlay topology

3 3 construction/maintenance is unertaken by low level mechanisms which in most of the cases are centralize or semi-centralize. Inicatively, such mechanisms are use in the Gnutella network [24], in which topology creation may be achieve by using a pre-efine aress-list of working noes inclue within a compliant client or by using web caches of known noes, a.k.a. Gnutella web caches. Similarly, Chor pre-assumes that noes are orere in a ring an are aware of their successor an preecessor in the overlay ring topology. Chor also relies on unerlying mechanisms for the overlay network bootstrapping [21]. In conclusion, p2p protocols are able to react to topology changes (an automatically re-assign key-value pairs) but are not responsible to create an maintain the overlay topology. Due to the ynamic nature of the MANETs, centralize or static mechanisms cannot be use for the bootstrapping of an overlay network. Thus, in the propose approach we hanle issues relate with topology formulation an maintenance in MANETs, where the existence of a stable network topology is not consiere as grante. 2.3 Routing protocols an topology formulation mechanisms in MANETs. Given the existence of a bootstrappe network, several mechanisms have been propose for the maintenance of the overlay topology in MANETs. These mechanisms vary from gossiping techniques [25] to exhaustive techniques [26]. The common characteristic of these mechanisms is that they pre-assume guarantee communication among the network noes. However their principles are completely ifferent. Gossiping techniques attempt to ientify the relative position of one noe in the overlay topology by consulting ajacent noes in the overlay topology. For these techniques the key inicator is convergence i.e. how many messages are require so as all noes know their relative position to the overlay topology. Alternatively, exhaustive techniques attempt to pass through all noes perioically in orer to ientify their relative position in the overlay topology. Since the topology formulation mechanism prerequisite the communication capability among the network noes, very crucial is the selection of the routing protocol that will be applie in the MANET. In MANETs there are no eicate routers. Multi-hop communication is establishe among neighboring noes an packets are transmitte from a source noe to a estination noe. Several routing protocols have been propose that perform variously epening on the type of traffic, the number of noes, the rate of mobility, etc. These protocols can be classifie in three categories: (i) proactive routing protocols where the routes to all the estination are etermine at the start up, an maintaine by using a perioic route upate process, (ii) reactive routing protocols where routes are etermine when they are require accoring to a route iscovery process an (iii) hybri routing protocols that combine both proactive an reactive protocol properties to come up with a better routing scheme. Accoring to the network topology an the network characteristics, the most suitable routing protocol may be selecte [ ]. Reactive routing protocols are more suitable for MANETs ue to their ability to cope with rapily changing topologies. However, they behave worse than proactive routing protocols in terms of scalability. In our approach, a reactive routing protocol is selecte but specific moifications are propose in orer to hanle scalability issues. 3. Autonomic approach for services provision 3.1 Generic principles The propose layere-approach aims at the provision of a generic framework that will facilitate the esign an evelopment of autonomic an ecentralize services in MANETs (see Figure 1). The introuction of the ifferent layers of the propose approach is necessary ue to the nee to aress the following challenges: a) efficiently utilize available network resources in a ynamic environment, b) provie services inepenently from the unerlying topology, c) ensure reliability of services in case of network topology changes an ) reuce the management complexity an increase flexibility to application evelopers. In orer to aress these challenges, autonomic functionalities have to be incorporate. The following self-* properties have been efine [6] an shoul be supporte by an autonomic system: self-configuration, self-optimization, selfawareness an self- healing. Fig. 1 Autonomic Services in MANETs

4 4 Existing protocols that satisfy partially the challenges escribe above were consiere uring the esign of the propose approach. There is no existing work on how to combine existing protocols for achieving autonomic service provisioning an how ifferent protocols coul interact using preefine interfaces. Taking into account these consierations, the propose approach is focusing on a) efining concrete layering for enabling autonomic service provisioning in MANETs, b) specifying the iscrete functionality of each layer an the interfaces between them an c) resolving conflicts between existing protocols, specifically in the fiel of the overlay topology construction. The creation an maintenance of an overlay topology that logically interconnects all the participating noes in the physical network is critical in our approach. Any noe that connects to the a-hoc network has to join to the overlay network. The overlay network is formulate uring the topology stabilization phase in an autonomic manner an hies any etails of the unerlying physical infrastructure, e.g. link establishment or torn own, noe failures, noe mobility, etc. In case of multiple changes in the physical topology, the overlay network is able to aapt quickly to the new environment (re-stabilization). Furthermore, recovery from failures can be easily succeee base on information that is available in the network. All these tasks are realize without the intervention of the network aministrator. After the overlay network is establishe, participating noes are able to store an retrieve ata using typical p2p protocols. Every noe that wishes to store a key-value pair, or query a value base on a key, can achieve it by using a Distribute Hash Table (DHT) [18] that operates on-top of the overlay topology. In a similar way, several applications can be built taking uner consieration the existence of a high level API put(key,value) an get(key) that woul interact with a DHT protocol that operates ontop of a non-reliable MANET. Provie services are esigne base on the assumption of collaboration an issemination of information among the participating noes. These services can be fully ecentralize as ata an functionality is allocate in ifferent noes at the overlay network. Some functions may be elegate to more than one noes for higher reliability. In case of changes or failures, roles may be reassigne autonomously an performance guarantees may be assure for the services provision. 3.2 Description of the four-layere approach We propose a four-layere scheme base on the functionality requirements impose by the provie services an the unerlying physical networking environment. As shown in Figure 2, the following four layers are efine; i) Neighbor-to-Neighbor layer, ii) Routing layer, iii) Topology Maintenance layer, an iv) DHT layer. Each layer has a iscrete role, implements ifferent mechanisms an specifies its messages types. The propose layere approach is inepenent from the selection of p2p protocols, topology formulation mechanisms an routing protocols. Therefore, any combination of ifferent protocols may be selecte an proper aaptations may be propose. Fig. 2 Four-Layere Approach The Neighbor-to-Neighbor layer is responsible for elivering an upper-layer frame from a neighbor to another neighbor. No information from the upper layer is necessary for the elivery. Two types of messages are use; i) MAC_SEND in orer to achieve one way frame elivery from neighbor X to neighbor Y an ii) MAC_ACK in orer to achieve acknowlegment for successful message-elivery from neighbor Y back to neighbor X. Also, this layer is responsible for maintaining (i.e. initializing an keeping up-to-ate) the routing cache of the Routing layer since when neighborto-neighbor links are create or estroye the relate routing information has to be upate. The Routing layer is responsible for elivering an upperlayer frame from a noe X to another noe Z. It is assume that noe X is not aware how noe Z can be reache. The layer is also agnostic of the reason that noe X wants to communicate with noe Z. This layer relies on routing protocol for frame forwaring across the network. As we state in section 2.3, in case of MANETs it is suggeste the use of a re-active routing protocol. The Topology Maintenance layer is responsible for formulating a virtual topology of the participating noes. In our case the esire topology is a ring (impose by the use of Chor). Consequently, this layer unertakes the task of ientifying the relative position of each noe in the overlay topology without being base in centralize or semi-centralize techniques. The DHT layer is responsible for maintaining a istribute hash table that is bootstrappe over the stabilize overlay topology. For this purpose any existing p2p protocol may be use. These protocols are (semi or fully) ecentralize an -in aition to storage an retrieval functionality- may succee loa balancing,

5 5 reuce banwith consumption an improve ata reliability across the network. The following interfaces have been efine for the communication among the ifferent layers: The Neighbor-to-Neighbor layer provies to the Routing layer routing information for existing neighbors that is store in the routing cache of each noe, through the valiateroutingcaches() function. The Neighbor-to-Neighbor layer provies also meium-level acknowlegments to the Routing layer for neighbor-to-neighbor communication, through the transfer_packet() function. The Routing layer provies routing functionality to upper layers through the routepacket() function. Aitionally, it exposes topology information erive irectly from the routing caches to the Topology Maintenance layer, through the getroutinginfo() function. It is up to the Topology Maintenance layer to utilize this information for optimizing its mechanisms or not. The Topology Maintenance layer provies information to the DHT layer regaring the relative position of a noe in the overlay network (e.g. the preecessor an successor in case of a ring topology) through the getrelativeposition() function. In case of changes in the network topology, stabilization proceures take place in both layers. The Topology_Stabilize() function is use for re-orering the overlay topology (e.g. ring in our case) an triggers the DHT_Stabilize() function that is use for the re-assignment of key-value pairs that are assigne in the overlay network noes. Fig. 3 Overlay Topology stabilization & DHT entries stabilization In Figure 3, a snapshot of the physical network topology (soli lines) an the logical overlay topology (ashe lines with arrows) is epicte. Initially, noe 3 oes not exist in the network an the key-value pairs have alreay been assigne to the network noes by applications that run on the existing noes (i.e. applications that use DHT). Then, noe 3 is physically connecte with noe 1 an noe 4 an the corresponing overlay topology is upate. It is the responsibility of Topology Maintenance layer to fin the successor for each noe. However, it is not the Topology Maintenance layer s responsibility to re-assign key-values accoring to the DHT s assignment algorithm. The Topology Maintenance layer must inform the DHT layer that the relative position for the noe in the overlay topology (e.g. ring in case of Chor) has change. Then it is up to DHT layer to reassign keyvalue pairs. This re-assignment will be aresse as DHT re-stabilization while the upate knowlege for the relative position in the overlay topology is calle Topology stabilization. 4. Implementation Details In this section, a reference implementation of the propose approach is escribe in etail. Specific mechanisms an protocols are selecte for each inepenent layer. In the DHT layer we have selecte Chor [26] as a p2p protocol, in the Topology Maintenance layer we follow gossiping an exhaustive tree-base techniques, in the Routing layer we have aapte the Data Source Routing (DSR) [30] protocol an in the Neighbor-to-Neighbor layer communication is establishe through a simulation environment. It is important to note that, in aition to aopting specific techniques an mechanisms, several aaptations - that were consiere useful for our approach - are provie. Specifically we escribe in etail how gossiping techniques for topology formulation an the DSR routing protocol can facilitate each other. These aaptations are state in etail in the following subsections Aapte DSR Routing Protocol Prior to analyzing the overlay topology algorithms that have been implemente we briefly refer to the reactive routing protocol that has been aopte an customize. The Data Source Routing (DSR) [30] is chosen for our reference implementation. Firstly, we escribe the basic characteristics of DSR an the implementation etails an seconly the propose aaptations that are correlate with the Topology Maintenance layer. DSR is relie in two mechanisms; a) Route Ientification an b) Route Maintenance. The protocol if fully reactive; hence every time one noe wants to communicate with another noe, it initiates a route request mechanism. Note that it is not DSR s responsibility to know why one noe wants to communicate with another noe. In general, this is upper layer s responsibility (Topology Maintenance layer in our case). The Route Maintenance mechanism is use for route ientification (see Figure 4). A specific message, calle DSR_Route_Request, is broacaste. Each DSR_Route_Request contains a request-i that is given by the initiator an the esire-estination noe. This information remains intact all across the flooing proceure. The message also contains a heaer that logs the routing path of a route-request. Thus, a route-request message that is initiate by one noe ens up to many route-request messages with the same request-i but with

6 6 ifferent heaers (because ifferent pathways are followe). Each noe that receives a DSR_Route_Request can perform three tasks: If a noe is the estination-noe it creates a route-reply message, calle DSR_Route_Reply. This message, in parallel with the request, contains a reply-i (which is irectly correlate with the request-i) an a heaer (which is actually the reverse route-request heaer), that remain intact until the reply reaches its estination. In case that a route-request reaches a noe that knows a-priori the route to the estination noe, it appens to the heaer the remaining route an creates a DSR_Route_Reply. If a noe knows nothing about the estination noe, it just forwars the DSR_Route_Request to its neighbors (excluing the neighbor that initially sent the request). In both replying an forwaring cases the noe as the request-i to a serve list. The serve list is a temporary list of all the request/response is that are serve by a specific noe, after the processing of a DSR_Route_Request (the same applies in routereplies). If a noe has the request-i of the DSR_Route_Request in its serve list, the noe oes nothing at all. This is very critical for avoiing loops. Fig. 4 Aapte DSR_Route_Request hanling mechanism When a DSR_Route_Reply reaches its estination, the noe that initially initiate the route-request is informe for the route to a specific noe. This information is store in the noes routing cache. Accoring to our implementation each noe maintains a primary routing cache an a seconary routing cache. The primary routing cache contains only one route for every known noe an specifically the route which is consiere to be optimum (for our implementation optimum means shortest but in general optimum may refer to other network parameters such as low latency etc.). The seconary routing cache contains multiple alternative routes without loops for known noes. In case that a broken link is ientifie (through DSR_Route_Error solicitation mechanism) both routing caches are revaliate. If an invali route -that contains this link- is ientifie in the primary routing cache, the route is remove an substitute by a possible alternative vali route, requeste from the seconary routing cache. As explaine above, the Routing layer is necessary so as two noes can exchange an upper layer frame. In orer to o so, a message calle DSR_Message_Transfer is use. When a route to a estination is known then a DSR_Message_Transfer is initiate. This message contains a packet-i (which is irectly correlate with the sener s aress), the estination noe, the route that must be followe in the MANET in orer to reach the estination an a flag that informs the estination noe whether it shoul respon with a confirmation (acknowlegement). Finally, the DSR_Message_Transfer contains encapsulate upper layer ata (topology formulation, DHT etc.). When a DSR_Message_Transfer has to be transmitte from noe A to noe B an from noe B to noe C, the intermeiate noes are responsibile to complete the transfer. Practically, if the DSR_Message_Transfer packet has reache noe B an the heaer of the DSR_Message_Transfer is A->B->C, assuming that noe B has irect connection with noe C (because in the past a route_reply informe noe A about that) an the connection between noe B an noe C is no longer available, then it is in the responsibility of noe B to ientify a new route to noe C. Ientification means usage of another route (consulting the routing cache) or initiation from scratch of a new route-request. Whenever uring a DSR_Message_Transfer a broken link is ientifie, then a message calle DSR_Route_Error is initiate by the noe that eclares the broken link. The DSR_Route_Error contains a routeerror-i an information about the broken link. This message is flooe all across the MANET. When a noe receives a DSR_Route_Error it automatically removes from the primary an seconary cache all the occurrences of routes that contain this link. Afterwars each noe re-evaluates its primary an seconary cache in orer for the optimum (shortest) paths to be pushe in the primary cache. At this point, we have to emphasize in some aaptations that we propose. The first aaptation regars the DSR_Route_Reply messages. Each DSR_Route_Reply contains a vali up-to-ate path that is route from the estination back to the route-request initiator, information that is extremely valuable also for the intermeiate noes. In our implementation all intermeiate noes overhear the route-replies that pass through them an enrich their routing cache. The secon aaptation regars the introuction of the Hopes-To-Live

7 7 (HTL) parameter. As we alreay mentione, the routerequest mechanism uses the request-i in orer to prevent infinite loops. The HTL parameter is ae as a fiel in the DSR_Route_Request. Every time a routerequest is forware then the HTL is increase by one. When the maximum HTL is reache the noe oes not forwar the DSR_Route_Request messages any more. The iea behin the HTL is the following; it is useless for a DSR_Route_Request to be forware more times than the iameter (in hops) of the network. As a iameter we efine the maximum non cyclic route that can be accomplishe. Of course, the problem for a MANET is that it oes not have a stable iameter an even if the noes are stationary the iameter has to be collaboratively inferre by a respective protocol. In our implementation we take in account the worst case as we state in the following paragraphs. network size for several network egrees. As we are going to see in the simulation results the introuction of the maximum HTL in the DSR, results in raical reuction of the DSR_Route_Request messages. It is important to note that, in orer to estimate the current epth of the network, base on the network size an egree, averaging techniques may be applie [33]. In these techniques, each noe interacts with its neighbours in orer to calculate the mean value of a parameter an convergence is succeee after a small number of iterations. Fig. 6 Virtual tree epth vs network size 4.2 Aapte T-MAN Protocol Fig. 5 Tree-base representation We assume that we have a network where each noe has approximately N neighbors ( refers to Network ensity). Without loss of generality we can represent this network as a tree with tree-egree T =N -1. As someone can notice from Figure 5, the worst case scenario for a route request is to go from one leaf to the root of the tree an back to another leaf; i.e. twice the tree epth. Consequently the maximum HTL that a route request must have in orer to fin one noe is tightly boun to the epth of the virtual tree that represents our network. We refer to this epth as k an we ientify the correlation of k with the network size N an the network egree N. The number of leafs of a tree with egree T an epth k is T k. The total amount of tree noes (or network noes) is N an the epth of the virtual tree is k, as it is shown in the following equations: N = N k T 0 = = + T ( N log( T ( 1) N N ( N log( N k + 1 ) 2 k = T 1 (1) ( k + 1 ) T 2 ) + 1 ) 1 ) ( 2 In Figure 6, it is shown that the epth of the virtual tree that represents our network is correlate with the ) Up to now we have examine the Routing layer that is the cornerstone regaring message exchanges for the upper layers. Such an upper layer is the Topology Maintenance layer. The topology formulation mechanism is responsible for ientifying the relative position of a noe among all the network noes. This is essential for proviing DHT functionality. As state earlier, the challenge is to apply a topology formulation algorithm for topology creation an maintenance to a MANET environment. One of the most prominent algorithms that have been propose for general purpose topology formulation is T- MAN [32]. The T-MAN approach is a gossip-base approach where each noe refines its view about its relative position in the overlay base on the knowlege of the closest noes that exist in a buffer. Each noe that participates in a network uses a ranking function to evaluate its istance from another noe. The evaluation parameters can vary. Some inicative examples are the network aress, proximity, latency etc. The usage of ifferent ranking function results in the creation of ifferent topologies. In our implementation we are base on the creation of an overlay ring topology. In orer to simplify our experiments we use as evaluation parameter the network aress an as a istance function: (a, b) = min(n a b, a b ).

8 8 scoring function to all the elements of the argument-list an returns only the first n occurrences where n is the size of the view. Table 2 Aapte T-MAN parallel threas Fig. 7 T-MAN stabilization process For example, we consier the topology where we have a network with 5 noes an we assume that the network aresses of the participating noes are 1,2,3,4 an 5 respectively (see Figure 7(a)). For the 3 r noe, noes 2 an 4 have score 1, while noes 1 an 5 have score 2. In orer to succee successor ientification as fast as possible, T-MAN applies a gossiping technique. Specifically, each noe maintains a view with the noes that are, up to a specific time, known an score. Perioically, each noe communicates with the most closer noe, it solicits its view an requests the current view of the closer noe (see Figure 7(b)). After this mutual exchange, noes re-evaluate their views ((see Figures 7(c) an 7()). This iterative proceure leas to extremely fast convergence (or stabilization), where convergence refers to the state that each noe knows its successor an any further exchange of messages leas to no further refinement of the views (see Figure 7(e)). Table 1 T-MAN parallel threas o forever{ Noe_To_Sent p selectclosernoe() buffer merge(view,{mynoedescriptor}) sen buffer to Noe_To_Sent p receive buffer p from Noe_To_Sent p buffer merge(buffer p,view) view Reevaluate(buffer) } active threa o forever{ receive buffer q from Sener q buffer merge(view,{mynoedescriptor}) sen buffer to Sener q buffer merge(buffer q,view) view Reevaluate (buffer) } passive threa T-MAN is relie on two parallel threas as it is shown on Table 1. Since T-MAN is propose for route networks, the only variable parameters that exist in the above Table are the size of the view an the size of the buffer. The operation sen buffer to Noe_To_Sent p of the active threa an sen buffer to Sener q of the passive threa in our layere approach is equivalent of initiating a DSR_Message_Transfer between these noes. As mentione above, a DSR_Message_Transfer consults the routing cache to fin out if a route to the estination exists. If not, the route-request mechanism is initiate. Metho merge returns a list that contains all the elements that exist in the sub-lists that are passe as arguments. Finally, metho Revaluate applies the o forever{ buffer merge(view,{mynoedescriptor}) routingcachelist ExtractNoesFromCache() buffer merge(view, routingcachelist) for (i=0;i<gossipfactor;i++){ Noe_To_Sent p selectclosernoe(i) DSRMessageTransfer to Noe_To_Sent p receive buffer p from Noe_To_Sent p buffer merge(buffer p,view) view Reevaluate(buffer) }} active threa o forever{ receive buffer q from Sener q buffer merge(view,{mynoedescriptor}) DSRMessageTransfer to Sener q buffer merge(buer q,view) view Reevaluate (buer) } passive threa Taking uner consieration that DSR is continuously working an contains up-to-ate routing information, an aaptation (aapte T-MAN) of the initial T-MAN is propose as it is shown in Table 2. In the aapte T- MAN approach there are the three following major changes: Before the active threa sens its view to the closest noe(s) it consults the DSR cache. Consulting means actually that all noes that exist in the primary an seconary cache are score (using the Ring-scoring function in our case). This consultation results in major reuction of the T-MAN messages that have to be exchange (before the ring is stabilize). Sen-to-noe is substitute by DSR_Message _Transfer as explaine above. Per each cycle, instea of sening one message to the closest noe that is in the noe s current view, we sen multiple messages (as efine by the parameter gossip factor). Multiple messages facilitate the fast issemination of information regaring the network topology, something that is very important in case of ynamic networks. It is obvious that the gossip factor can be between one an buffer size. Proper selection of this option may accelerate the convergence while keeping the routing overhea low. 4.3 Exhaustive tree-base approach In orer to have a comparative view of T-MAN we have also implemente an exhaustive tree-base approach (see Figure 8). Accoring to this approach each noe of the network is responsible to fin its successor by using an excessive flooing mechanism. Each noe in the network solicits a Fin_Successor_Request message after assigning a request-i to that message (see Figure 9a an 9b). This message flows across the noes using a Neighbor-to-Neighbor communication scheme (an not DSR noe-to-noe message transfer scheme as it

9 9 happens at aopte T-MAN). After sening the message a serve list is enriche in each noe with the request i. In this way each noe keeps track of the request-is that have been forware by the noe. Before the Fin_Successor_Request is forware to the next neighbor, each noe enriches its heaer with all the noes that are visible as neighbors to that noe. Consequently when the next noe receives the Fin_Successor_Request, it has to check if all of its neighbors exist in the heaer of the request. If not, it enriches the heaer an forwars the message to the neighbors that i not exist in the heaer (see Figure 9c an 9). The critical part of the process is that if one noe receives a Fin_Successor_Request an the request-i is not in the serve list an the heaer contains all of its current neighbors, then this noe is consiere to be a virtual leaf of a tree that has as root noe the initiator of the Fin_Successor_Request (see Figure 9e). The number of Fin_Successor_Responses in orer to succee stabilization of the overlay ring topology is the sum of the Fin_Successor_Responses that receives each noe in the tree. The root noe receives T k responses (the number of the virtual leafs). In the first level of the k tree there are T chilren where each one receives T 2 responses. In the secon level of the tree there are T chilren where each one receives T k responses. In the (k-1) level of the tree there are T k-1 chilren where each noe receives T k responses. Finally, in the k level of the tree there are T k chilren where each noe receives T k k *(T -1) responses. Thus, the number of Fin_Successor_Responses that are require for stabilization is: (T Routing_Cost = k -1)T T -1 ( N Routing _ Cost = (k+ 1) (2 1) N (( N = + 1) N 2 k k 1) 1)( N 1) N (3) ( k + 1) 4.4 Chor p2p Protocol Fig. 8 Exchaustive tree-base flooing mechanism Fig. 9 Tree-base stabilization process The virtual leaf uses DSR s message transfer mechanism to sen a Fin_Successor_Response back to the initiator (root). This message contains the request-i of the Fin_Successor_Request an the heaer that was graually built up to the point that a noe inferre that is a leaf. Consequently a Fin_Successor_Request initiate by a noe will result in as many Fin_Successor_Responses as the number of the virtual leafs that exist. Also DSR s message transfers are equal to the amount of Fin_Successor_Responses. Each root that receives the Fin_Successor_Responses upates its view as far as its successor is concerne. In orer to maintain the DHT on top of the stabilize ring topology, we have selecte Chor [21]. Chor is a simple but powerful protocol, which solves the problem of efficient ata storage an retrieval. It is an efficient istribute lookup system base on consistent hashing. Its only operation is to map a key to a responsible noe. Each noe maintains routing information about O(log N) other noes, an lookups are feasible via O(log N) messages. Therefore, Chor scales well with a number of noes an, thus, it can be applicable to large systems. Chor continues to function correctly even if the system unergoes major changes or if the routing information is partially correct [21]. Chor oes not implement services irectly but rather provies a flexible, high-performance lookup primitive, upon which such functionality may be efficiently layere. Its esign philosophy is to separate the lookup problem from aitional functionality. By layering aitional features on top of a core lookup service, overall systems will gain robustness an scalability [22]. The DHT layer uses three types of messages; i) PUT(k,v) in orer to associate the value v with the key k, ii) GET(k) in orer to retrieve the value that is associate with the key k an, iii) STABILISE in orer to move parts of the keys that are store in the local cache to another noe ue to changes in the ring (refer also to section 3.2). 5. Experimental Evaluation In this section we investigate the behavior of the propose implementation, as it is escribe in Section 4. Each layer that was previously presente -except Neighbor-to-Neighbor layer- was implemente separately in PeerSim [31]. We assesse the performance

10 10 of each layer in terms of messages exchange, generate errors an convergence capability in iverse network sizes. Each simulation was execute three times an average values were consiere in our analysis. 5.1 Topology an Simulation Results In the next scenarios, multiple noes are simultaneously activate in the simulate a hoc network. Key parameters of the topology are the number of noes (N) (size of the network) an the average number of neighbors () (ensity egree). In all simulations, any operational noe automatically recognizes its neighbors an joins the overlay network. Each noe keeps state relate to its successor in the ring topology. In case of noe failures, establishe connections are torn own an re-stabilization proceures take place. Initially, we compare the performance of the propose aaptations in terms of total number of messages exchange until the overlay ring topology is formulate. Simulations are taken for practically size an egree (ensity) networks. The Hops-To-Live parameter in DSR protocol was set to infinite an optimum respectively. Fig. 11 Aapte T-MAN vs tree-base messages In Figure 11, we compare the performance of aapte T- MAN an exhaustive tree-base approach. It can be shown that the number of messages for topology formulation in the tree-base approach is significant larger compare to the aapte T-MAN. This is ue to the fact that in the tree-base approach, exhaustive neighbor to neighbor communication takes place an also that in the aapte T-MAN approach the information that is available in the noes routing caches is exploite. Fig. 10 Aapte DSR vs DSR messages In Figure 10, the performance of the propose approach with the use of the aapte DSR compare to the pure DSR is epicte. It is evient that the use of pure DSR unerperforms significantly compare to the aapte DSR. This is reasonable, since the exploitation of routing cache information generate by DSR in aapte T- MAN makes easier the ientification of the successor noes in topology formulation phase an thus requires less messages. At this point, it is important to note that the selection of T-MAN instea of aapte T-MAN was not consiere a priori as a viable solution, since T-MAN is not able to converge faster than the aapte T-MAN. This is ue to the fact that T-MAN oes not consult the existing routing caches of the noes an therefore more cycles are require for overlay topology stabilization [32]. Fig. 12 Aapte T-MAN performance with HTL infinite These simulation results le us to focus on the aapte T- MAN algorithm an further investigate the impact of other parameters, such as the routing HTL parameter or the network egree in the overlay topology formulation an maintenance process. In Figure 12 we can observe that as the number of noes in the network increases, the number of aapte T-MAN messages also linearly increases. In aition, fewer messages are require for topology formulation in ense networks (high egree) as each noe exchanges more precise gossiping messages. Precision is achieve ue to the scoring of larger number of neighbors. Finally, the number of messages remains almost stable in large egree networks.

11 affects the flooing of routing messages in the network an, thus, the routing overhea is significantly reuce by approximately an orer of magnitue. 11 Fig. 13 Aapte T-MAN performance with HTL optimum In Figure 13, the effect of the Hops-To-Live (HTL) parameter to the total number of aapte T-MAN messages is epicte. In case the HTL parameter is set to the optimal value accoring to the Equation 2 in section 4.1, the number of T-MAN messages is slightly increase compare to the previous case. As expecte, by limiting the HTL parameter we reuce the flooing of routing messages across the network -to the absolute minimum for aressing routing nees- an, thus, smaller routing caches in each noe are exploite by aapte T-MAN stabilization phase. Fig. 15 DSR Route Requests with HTL optimum Thus, we conclue that the effect of the HTL parameter is extremely important because, when the parameter is set to its theoretical optimum value, aapte T-MAN messages were slightly increase while in parallel the routing overhea was reuce by an orer of magnitue. Figure 16 epicts the number of aapte T-MAN messages for ifferent network sizes an variable ensity. An important observation is that the number of messages converges quickly for ifferent network ensity egrees. This observation coul potentially be exploite in builing clustering algorithms, where the network ensity coul be use to estimate the best clustering size. Fig. 14 DSR Route Requests with HTL infinite Furthermore, we aim to observe the amount of DSR s route-request messages that is exchange until the overlay ring topology is formulate. In Figure 14 an 15, the HTL parameter is set to infinite an to optimum respectively. As shown in both figures, more routerequest messages are generate as the size of the network increases. This is expecte as DSR messages are generate in response to T-MAN messages, which are analogous to the number of noes (Figure 12 & 13). Furthermore, sparse networks require more routing messages than ense networks, especially as the number of noes increase ue to smaller number of entries generate via neighbor-exchanges. The aapte T-MAN Fig. 16 Aapte T-MAN message for various sizes an ensities 5.2 Inicative Service - Visualization Upon stabilization of the overlay ring, a visualization service is provie. Each noe stores topology ata that is available through information acquire from neighbor iscovery messages an also from information existing in the routing cache of each noe. All noes use a preefine key (e.g. network_topology ) in orer to

12 12 ientify the responsible noe for the storage of the visualization ata. The visualization cost, i.e. the total number of messages exchange for the visualization ata to be store, is calculate. Fig. 18 DHT_PUT failures 6. Conclusions an Future Work Fig. 17Visualization Cost Figure 17 epicts the number of aapte T-MAN an DHT_PUT messages for constant an variable network ensity. In the first case, the network expanses without any change in its ensity while in the later case the ensity is proportional increase as the network sizes increases. We notice that the number of DHT_PUT messages require for the provision of the visualization service, is slightly affecte by the network size an ensity. It is important to note that in case that the propose aaptations in the T-MAN an DSR algorithm were not applie, the visualization cost woul be significantly higher, as alreay shown in Figure 10. In the last set of simulations, we investigate whether the visualization service is robust to multiple noe failures. Uner the unfavorable conition that 20% - 30% of network noes become simultaneously non-operational an for various network sizes, we measure the number of DHT put(key, value) calls that generate DSR_Route_Error messages for any broken link ientifie. Such messages are flooe across the network in orer to upate the routing cache for future requests (see Section 4.1). As shown in Figure 18, the majority of the put(key, value) requests are successfully complete without any elay (error). For example, for a 40-noe network an 30% noe failure, 17 messages were successful while another 10 faile. Even if a put(key, value) request initially fails ue to routing inconsistencies, the requests will be eventually be successful after the routing tables are upate. Therefore, there is only a time penalty for the service elivery after a major network failure. It coul be argue, therefore, that the visualization service remains operational even when a significant portion of noes (an attache links) fails. Similar trens shoul be notice for other istribute services, which have to be verifie with further simulation experiments. In this paper we presente a new paraigm for service evelopment an operation in mobile a hoc networks. We initially propose a four-layer approach for setting up an overlay network, over which services are eploye. Each layer may be inepenently realize with existing peer-to-peer an routing protocols. Topology formulation, maintenance an routing challenges in MANETs are aresse by suggeste aaptation of existing algorithms, such as T-MAN an DSR. It coul be argue that the overlay network hies from the service layer any physical network topology changes an, thus, service evelopment is significantly simplifie. Using extene simulations, we showe that the aapte algorithms quickly converge while the number of messages is kept low. In a reference implementation, we showe that a istribute network service coul efficiently run over a ynamic network. We argue that other services may also be implemente base on our approach. In our future work, these initial conclusions will guie the evelopment of a more etaile prototype, possible taking in to consieration functionality of IPv6 protocols. Acknowlegments This publication is base on work partially performe in the framework of the European Commission ICT/FP7 project EFIPSANS ( References [1] A. Galis, et al, Management an Service-aware Networking Architectures (MANA) for Future Internet, Communications an Networking in China (ChinaCOM), pp.1-13, Aug [2] R. Bless, C. Hiibsch, S. Mies an O.P. Walhorst, The Unerlay Abstraction in the Spontaneous Virtual Networks (SpoVNet) Architecture, Next Generation Internet Networks, NGI 2008, pp , April, 2008.

13 13 [3] A. Bassi, S. Denazis, A. Galis, C. Fahy, M. Serrano, J. Serrat, Autonomic Internet: A Perspective for Future Internet Services Base on Autonomic Principles, In: IEEE 3r Intl. Week on Management of Networks an Services Manweek 2007 / MACE [4] K. Tutschkua, P. Tran-Giaa an F.U. Anersenb, Trens in network an service operation for the emerging future Internet, International Journal of Electronics an Communications, vol.62, pp , October [5] G. Kunzmann, A. Binzenhöfer, A. Stäber, Structure Overlay Networks as an Enabler for Future Internet Services, it - Information Technology, vol.50, pp , no.6, November [6] B. Jennings, S. Van er Meer, S. Balasubramaniam, D. Botvich, M.O. Foghlu, W. Donnelly an J. Strassner, Towars Autonomic Management of Communications Networks, IEEE Communications Magazine, vol.45, no.10, pp , October [7] A. Greenberg, G. Hjalmtysson, D.A. Maltz, A. Myers, J. Rexfor, G. Xie, H. Yan, J. Zhan an H. Zhang, A Clean Slate 4D Approach to Network Control an Management, ACM SIGCOMM Computer Communication Review, vol.35, no.5, pp.41 54, [8] H. Ballani an P. Francis, CONman: A Step Towars Network Manageability, ACM SIGCOMM Computer Communication Review, vol.37, no.4, pp , [9] D. Clark, C. Partrige, J.C. Ramming an J.T. Worclawski, A Knowlege Plane for the Internet, Proc. of the 2003 conference on Applications, Technologies, Architectures an Protocols for Computer Communications, pp.3 10, [10] M. Caleffi an L. Paura, P2P over MANET: Inirect Tree-base Routing, IEEE International Conference on Pervasive Computing an Communications, [11] R. Chaparaza, Requirements for a Generic Autonomic Network Architecture Suitable Requirements for Autonomic Behavior Specifications of Decision-Making-Elements for Diverse Networking Environments, International Engineering Consortium (IEC) Annual Review in Communications, vol.61, December [12] G. Diaz, Autonomic Networks an Management of Dynamic Services Deployment - A Stuy of Approaches by Using Overlay Networks, IEEE Global Telecommunications Conference, [13] A. Klenk, M. Kleis, B. Raier, S. Elmoumouhi, G. Carle an M. Salaun, Towars Autonomic Service Control In Next Generation Networks, Fourth International Conference on Autonomic an Autonomous Systems, pp , March [14] B. Biskupski, J. Dowling an J. Sacha, Properties an Mechanisms of Self-Organising MANET an P2P Systems, ACM Transactions on Autonomous an Aaptive Systems (TAAS), vol.2, issue 1, [15] D. Hora an D. Maceo., Enhancing peer-to-peer content iscovery techniques over mobile a hoc networks, Computer Communications, vol.32, issue 13-14, pp , August [16] M. Bisignano, G. Di Moica, O. Tomarchio an L. Vita, P2P over Manet: a comparison of cross-layer approaches, 18th International Conference on Database an Expert Systems Applications, DEXA '07, pp , 3-7 Sept [17] Y. C. Hu, S. M. Das an H. Pucha, Peer-to-peer overlay abstractions in manets, Theoretical an Algorithmic Aspects of Sensor, A Hoc Wireless an Peer-to-Peer Networks, CRC Press, [18] E. Kang et al, DHT-Base Mobile Service Discovery Protocol for Mobile A Hoc Networks, Proceeings of the 4th international conference on Intelligent Computing, pp , [19] T. Wongsaarsakul an K. Kanchanasut, A Structure Mesh Overlay Network for P2P Applications on Mobile A Hoc Networks, Distribute Computing an Internet Technology, ISBN: , [20] M. Jing-qing et al, Cross-layer optimize Chor protocol for separate ring convergence in MANET, The Journal of China Universities of Posts an Telecommunications, Elsevier B.V., [21] I. Stoica, R. Morris, D. Karger, M.F. Kaashoek an H. Balakrishnan, Chor: A Scalable Peer-to-peer Lookup Service for Internet Applications, In Proceeings of SIGCOMM 2001, San Deigo, CA, August [22] A. Rowstron an P. Druschel, Scalable, istribute object location an routing for large-scale peer-to-peer systems, In Proceeings of the 18th IFIP/ACM International Conference on Distribute Systems Platforms, Heielberg, Germany, November [23] E.K. Lua, J. Crowcroft, M. Pias, R. Sharma an S.Lim, A Survey an Comparison of Peer-to-Peer Overlay Network Schemes, IEEE Communications Survey an Tutorial, vol.7, no.2, pp.72-93, [24] Y. Wang, X. Yun, an Y. Li, Analyzing the Characteristics of Gnutella Overlays, Fourth International Conference on Information Technology, pp , April [25] R. Frieman an D. Gaviia, Gossiping on MANETs: the Beauty an the Beast, ACM SIGOPS Operating Systems Review, vol.41, issue 5, October [26] W. Ding an S.S Iyengar, Bootstrapping Chor over MANETs - All Roas Lea to Rome, IEEE Wireless Communications an Networking Conference, [27] S. Zaki, M. Ngai an S. Razak, A Review of Delay Aware Routing Protocols in MANET, Computer Science Letters, vol.1, [28] M. Abolhasan, T. Wysocki an E. Dutkiewicz, A review of routing protocols for mobile a hoc networks, A Hoc Networks, vol.2, issue 1, pp.1-22, January [29] G. Jayakumar an G. Gopinath, A Hoc Mobile Wireless Networks Routing Protocols A Review, Journal of Computer Science, vol.3, no.8, pp , [30] D.B. Johnson an D.A. Maltz, Dynamic Source Routing in A Hoc Wireless Networks, Kluwer Acaemic, [31] M. Jelacity, A. Montresor, G. Jesi an S. Voulgaris, The Peersim simulator, accesse March [32] M. Jelasity an O. Babaoglu, T-Man: Gossip-Base Overlay Topology Management, In Proceeings of Engineering Self- Organising Systems, ISBN: , July [33] M. Jelasity, A. Montresor an O.Babaoglu, Gossip-base aggregation in large ynamic networks, ACM Transactions on Computer Systems, vol.23, no.3, pp , Panagiotis Gouvas receive the Dipl.-Ing egree in Electrical an Computer Engineering from NTUA, Greece in He is currently a PhD Caniate in the Information Management Unit in NTUA. He was involve in technical an project management aspects of multiple national an European Commission research projects. Anastasios Zafeiropoulos receive the Dipl.-Ing egree in Electrical an Computer Engineering from NTUA, Greece in He is currently a network engineer in GRNET S.A. an a PhD Caniate in the Multimeia Communications an Web Technologies Research Group in NTUA. He was involve in technical an project management aspects of multiple national an European Commission research projects.

14 14 Athanassios Liakopoulos receive the Dipl.-Ing. egree in Electrical an Computer Engineering from the NTUA, Greece in 1996, MSc with Distinction in Telematics from the Electrical Engineering Department in University of Surrey (UniS) in 1998 an PhD in Electrical & Computer Engineering from the NTUA in Since September 2000, he joine the GRNET S.A. where he is Coorinator of Networking an Computing Infrastructure. Dr. Gregoris Mentzas is Professor of Management Information Systems at the School of Electrical an Computer Engineering of the NTUA an Director of the Information Management Unit (IMU), a multiisciplinary research unit at the University. His area of expertise is information technology management an his research concerns the integration of knowlege management, semantic web an e-service technologies. Prof Nikolas Mitrou s research interests are in the areas of igital communication systems & networks an networke multimeia in all range of stuies: esign, implementation, moelling, performance evaluation an optimization. Since 1988 Prof. Mitrou has been actively involve in many RACE, ACTS an ESPRIT projects an he was the coorinator of one of them (AC235, WATT). He is a member of the IEEE, member of the IFIP WG 6.3 an member of the Technical Chamber of Greece.