(4)Routing in MANET and P2P Nra Comparison

8/8/2019 (4)Routing in MANET and P2P Nra Comparison

http://slidepdf.com/reader/full/4routing-in-manet-and-p2p-nra-comparison 1/15

Routing in Mobile Ad Hoc and Peer-to-Peer Networks.

A Comparison

Rüdiger Schollmeier 1), Ingo Gruber 1) and Michael Finkenzeller 2)

1)Lehrstuhl für Kommunikationsnetze, Technische Universität München,Arcisstr. 21, 80333 München, Germany

{Schollmeier, Gruber}@lkn.ei.tum.de2)Siemens Corporate Technology, Siemens AGOtto-Hahn-Ring 6, 81730 München, Germany

[email protected]

Abstract. The goal of this paper is to make the similarities and differences of Peer-to-Peer (P2P) and Mobile Ad Hoc (MANET) networks clear. Thus we want to show the synergetic

potential hidden in these two decentralized and self organizing networks. Peer-to-Peer andmobile ad hoc networks are established on a different basis. Peer-to-Peer is based on an IPnetwork and Mobile Ad Hoc networks are based on a mobile radio network. However bothnetworks hold similarities concerning their routing and network management principles. Thereason therefore is, that both of them have to solve a similar goal, namely to providenetworking functionalities in a completely unmanaged and decentralized environment. One of the most interesting tasks in these networks is thus how queries are guided through the network.Therefore we concentrate in this work on the different routing algorithms employed in Peer-to-Peer and mobile ad hoc networks, and thus conclude the similarities and differences of bothnetworks. Finally, the similarities of both networks can be used, to bring up the synergeticeffects of looking at both networks at the same time, which is done at the end of this paper.

1. Introduction

Since the first appearance of wireless ad hoc networks as the DARPA packet radio networks in the1970s [12, 14], they became an interesting research object in the computer industry. During the lastcouple of years tremendous improvements are made in the research of ad hoc networks. Thewireless LAN standard 802.11 [23] is used as a wireless connection of portable computers with thelocal network. However it still does not supply completely self configuring ad hoc networking. Withthe development of Bluetooth [4] a first product, designated only for ad hoc networking, isavailable. Due to its possibility to create and organize a network without any central management,ad hoc networking is characterized as the art of networking without a network [6].

On the other hand, a similar concept without infrastructure can be observed in the Peer-to-Peer networking area. Peer-to-peer networks are first discussed in the mid 1990s, and became famous inthe late 1990s as file sharing platforms, especially to distribute mp3-compressed music tracks.

In this case the IP-layer provides the basic communication medium, which enables IP capableterminals to reach anyone and anything attached to the IP-network. However, just like the pure radiotransmission in ad hoc networks does not facilitate the search for connection-partners, the IP-layer does not tell a terminal how and where to find content or other participants. These peer-to-peer networks, like e.g. Freenet [13] or Gnutella [7] are completely self organizing networks.

Similarities between both networks arise, as the basic problem, how to enable bidirectionalterminal to terminal communication in an unmanaged environment, is the same. However, besidethe similarities there are also great differences, due to the different underlying network layer and themotivation for creating an ad hoc or peer-to-peer network.

As the routing architectures are one of the most important modules within an unmanagednetwork, the routing algorithms of these two kinds of networks, Peer-to-Peer and wireless ad hoc



networks will be compared in this work. As a result we want to point out similarities and differences between the Peer-to-Peer and the wireless ad-hoc networking world. Thus it could be possible to

make use of the synergetic effects which may occur if the solutions in these two networks with acompletely different physical layer but the same goal, namely to provide networking functionalitieswithout a given network, are compared.

The paper is structured as follows: a brief overview of the main attributes of wireless ad hocnetworks is given in Section 2. In Section 3 we present an overview about the main routingalgorithms in ad hoc networks, which is followed by a description of the characteristics of peer-to-

peer networks in Section 4. Section 5 then enfolds the routing algorithms in peer-to-peer networks,and their differences and similarities to ad hoc networks are then discussed in Section 6. FinallySection 7 concludes this work and gives an outlook to further possible developments and researchtopics in this area.

2. Ad Hoc Networks

Ad hoc networks are self configuring wireless networks. An ad hoc terminal is by definition asource of information, a drain for information and a router for information flows from other users. Awireless ad hoc network allows unrestricted mobility of the enlisted terminals, as long as at least oneterminal is within transmission range. Direct neighbors can be used as relay stations to connect tonodes beyond a terminals own coverage (see Figure 1 ). The transmission range of a wireless ad hocnode is restricted by its power consumption due to the limited battery power of a wireless node.Therefore the node transmission range is typically small compared to the span of the network.

All nodes within a wireless ad hoc network use the same frequency band as a shared medium for receiving and transmitting data. The used frequency band is an unreliable channel for datatransmissions. Unpredictable errors can occur on the transmission channel, and therefore a reliabledata transmission protocol and a forward error correction are mandatory to reliably send out packetsfrom a sender to a destination. A transmission of data from the sender to the receiver has to be done

by a broadcast over the shared medium. This might lead to collisions of packets due to overlappingtransmissions of two senders. Therefore, a wireless node first sense the physical medium for a giventime period ∆ t, before sending a packet. This is an improved collision avoidance algorithm, toreduce the possibility of packet loss due to collisions on the physical medium. Collision avoidancehelps to use the limited bandwidth in an optimal manner. But it does not solve the hidden terminal

problem. A hidden terminal situation happens, when a receiver gets packets from two senderssimultaneously but the senders are too far away from each other to sense, that the other node alreadyuses the shared medium to send packets to the receiver, and therefore most of the transmitted datacollides and is corrupted, and must be retransmitted by the senders.

An ad hoc network is mostly user driven, not data driven. The purpose of an ad hoc network is toconnect users with their terminals with other users, instead of redundantly distribute data over thenetwork. This means, that a route is setup from a source to a specified destination, which aredefined by their unique addresses, for the reason of enabling a communication between both users.

The advantage of this kind of network is, that it does not require or even need any kind of infrastructure, like a base station in a cellular network. Therefore ad hoc networks are best suited for an environment, which is not able to provide any kind of infrastructure, e.g. in a hostile environmentor for disaster recovery.

However the lack of infrastructure also has some disadvantages. One disadvantage is itsimplication on the security issue. This is an important issue, because a route from the source to adestination is routed by intermediate nodes and every intermediate node as well as every nodereceiving the transmissions between intermediate nodes can possibly intercept the communication.Therefore an encrypted end-to-end connection is important for a private communication. Whileencryption is easy to put into practice with the existing tools known from the Internet,authentication of communication partners is not easy or might even be impossible to guarantee



within wireless ad hoc networks. A public key infrastructure (PKI) as in the internet is not feasiblein wireless ad hoc networks.

Ad-hoc networks according to the definition outlined above, thus suffer from the lack of aninfrastructure and additionally from the terminal mobility. Due to the terminal mobility, the links

between neighboring nodes are dynamic. While two nodes are diverging, the link between these twonodes breaks, when the distance between both is too large, and they leave the radio range of eachother. If this is the case, routing tables, which contain the connected neighbors, must be updated atleast in these two nodes. These new informations must then be used for new connection requests. Asterminals are not only sources and drains of information, but also routers for others, terminals mustnotify all other nodes in the network, about this change in the topology of the network, therewiththey are enabled to compute the shortest path between two other nodes.

The breakup of connections between adjacent nodes leads to frequent topology updates, which allhave to be announced to all other nodes within the network. While the terminal mobility is low, andthe number of nodes in the network is small, updates must be sent seldom. However, if the velocityof the terminals, or the number of nodes is increasing, the topology updates are getting more

frequently, and the signaling messages are using more link bandwidth. Therefore ad hoc networksdo not scale well for large numbers of terminals, especially with high mobility patterns. Due to this

behavior of ad hoc networks, it is important to employ optimized routing algorithms to use thescarce available bandwidth in a most optimal pattern.

Figure 1 A simple ad hoc network: nodes havedifferent directions and velocities and a multi-hopad hoc connection is established (grey pointers)

Figure 2 Schematic network topology of a Peer-to-Peer network

3. Peer-to-Peer Networks

Since a few years Peer-to-Peer (P2P) networks came into discussion, which provide the capability toestablish virtual overlay networks. So called “pure” P2P networks which are completely self organizing and therefore do not need central instances to manage the network. In this work, our understanding is according to the understanding of Peer-to-Peer networks, as described in [22]. Themain characteristic of Peer-to-Peer networks is from our point of view, that the terminals of thesenetworks communicate in a bidirectional and symmetric way with each other. Therefore a virtualoverlay network is established above the IP-Layer (see Figure 2 ). Such a network consists in mostcases only of the servents and the TCP/IP connections between the different servents. The term servent



in this context represents the fact, that each participant in a Peer-to-Peer network acts as a server and asa client, as understood in the common sense, at the same time. Therefore the artificial word servent has

been created, which is constituted of the first syllable of the term server and the second syllable of theterm client.

The servents itself are mostly computers with standard computing capacities, nearly infinite power supply (as they are connected to the electricity network) and at least with a network connection of about 56 kbit/s. Thus the terminals in Peer-to-Peer networks are more powerful than the terminalsnormally used in a wireless environment.

The virtual network, based on the servents and the TCP/IP connections between them, is used tosatisfy the different demands of the users operating in this network, may it be the search for an mp3compressed audio file, or for another participant to whom a Voice over IP connection should beestablished. The data which is exchanged between the peers directly, may range from file data, over Internet Relay chat data up to real time voice communication and interaction data streams.

Basically there are no hierarchies given by the IP-Layer or the peers themselves. However this doesnot exclude the possibility to establish a routing architecture for the Peer-to-Peer network, which

brings in central entities and thus adds hierarchies to the network. These central entities may be used ascentral look up tables to route queries, or as a kind of dynamic proxy server for low-bandwidth clients(see Section 5.1). Depending on the way, central entities are added to the routing architecture, wedistinguish between several routing approaches, e.g. flat, hybrid and hierarchical, as described inSection 5.

However, employing central entities may on the one hand be of advantage because of managementand scalability issues. On the other hand central entities also impose new problems like a single pointof failure. Thus depending on the kind of application the Peer-to-Peer network is employed for, it must

be carefully decided, what kind of routing architecture is used.In contrast to wireless Ad-hoc networking, Peer-to-Peer networking is already widely accepted e.g.

in Napster [16] or Gnutella [7]. The cause therefore can be found in the easy availability andaccessibility of the IP-Layer, which makes the development of Peer-to-Peer networking terminalsmuch easier, than in the wireless world, with its additional physical constraints.

4. Routing in Ad Hoc Networks

Several different routing algorithms for ad hoc networks, with their special advantages anddisadvantages have been proposed until now. They can be divided in two main branches, the proactiveor table drive routing algorithms and the reactive or on demand routing algorithms. A node running a

proactive routing algorithm has the full network view at every time, like a regular router in theInternet. All topology updates are broadcasted immediately or with a small time shift to all other nodesin the network. Therefore the route establishment can take place very fast. The disadvantage of

proactive routing algorithms is the number of required topology updates within a time period. In casethe number of nodes belonging to a network rises over a certain threshold, this kind of routingalgorithm is not feasible anymore.

In contrast to that, nodes using a reactive routing algorithm do not send any kind of topologyupdates to its neighbors. Only in case they want to set up a route to another node, they flood a routerequest through the network, and get a response from the destination or an intermediate node, whichknows the route to the destination by a formerly made route request.

4.1. Destination Sequenced Distance Vector Routing (DSDV)

The Destination Sequenced Distance Vector Routing protocol [17] is the best known protocol for a proactive routing scheme. It is based on the classical Distributed Bellman Ford (DBF) routingmechanism [2]. The basic algorithm is optimized to guarantee the avoidance of loops in the routingtables. Like every other table driven routing algorithm it maintains several routing tables which

provide information about every possible destination within the network. The tables contain the



minimum number of hops and the next hop in direction of a specific destination. Further on, they alsoinclude sequence numbers for every destination. The sequence numbers guarantee that the node candistinguish between stale and new routes. Therefore it can update its routing tables with more actualinformation (higher sequence numbers), and thereby avoid route loops. Route updates can be sent astwo types. Either a full route update, where all information of all tables of the node are exchanged, or an incremental update to save link bandwidth. The incremental updates only contain information aboutchanges since the last full update.

Broadcasted route updates contain the address of the destination, the number of necessary hops toreach the destination, the sequence number of the route information regarding the destination, and anew sequence number unique to the broadcast. Receivers of this route updates can thus update their own routing tables, and if necessary broadcast their own routing tables. While every node maintains allnecessary information of the full network, a route setup can be processed very fast with the locallystored information. Unlike the reactive routing algorithms, no more flooding of a route request throughthe network is required to discover a route from the source to a destination.

4.2. Dynamic Source Routing (DSR)

The Dynamic Source Routing [11, 10] is one of the major on demand routing algorithms. It is based onthe concept of source routing. The protocol includes two major phases: route discovery and routemaintenance.

The route discovery starts when a mobile node tries to send a packet to a destination. In a first step,the node checks its own routing cache. In case the routing cache has a valid route to the destination,this route is used to forward the packet. In case the cache does not contain a route, it broadcasts a routerequest to its neighbors containing its own address, the destination address and a unique sequencenumber for loop detection. The receiving node checks its cache for a route to the destination. If it doesnot contain a route it adds its own address to the packet and forwards it. In case a node receivesmultiple copies of the same route request from different intermediate nodes, it only forwards the firstrequest, and discards the following copies of the request. A route reply will be sent back if a node has

an actual route to the destination, or the route request reaches the destination. If the node is not thedestination, it adds its cached route to the route reply message. In any case, the route reply messagecontains the full route from source to destination and exactly follows this route in reverse direction.Every data packet traversing the network from the source to the destination, contains the full path withthe addresses of all intermediate nodes. Therefore the intermediate nodes do not have to cache the pathand can save memory. This advantage of demanding less memory in the intermediate nodes is also adisadvantage, because the size of every packet is increased as it contains the full route information.

Further on, route error packets fulfill the task of route maintenance. Route error packets are sent back to the source when an intermediate node or the destination is not reachable anymore. Every nodeforwarding an error packet, deletes the broken hop and all routes relying on that broken hop. Thesource node also deletes the stale route and starts a new route request.

4.3. Ad hoc On-Demand Distance Vector Routing (AODV)

Ad hoc On-Demand Distance Vector Routing [19, 18] is also an on demand routing algorithm, but incontrast to DSR not source based routing, but every hop of a route maintains the next hop information

by its own.AODV uses, like DSR, route request (RREQ) and route reply (RREP) messages. The route request

packet is used, if a source wants to send a message to a destination. If its routing cache does notcontain a route to the destination, it broadcasts a RREQ to its neighbors. The RREQ contains thesource address, the destination address and a unique sequence number. Every node receiving a RREQsaves the node address, from which it first received the RREQ, before it forwards the message to itsneighbors. In case a node receives copies of the same RREQ, the copies are discarded. The RREQ isforwarded through the network until the RREQ reaches a node with a valid route to the destination, or the destination itself. The intermediate node or the destination creates a RREP message, and sends it to



the node from which it first received the RREQ. The RREP is forwarded by the nodes on the reverse path the RREQ traveled. Every node saves the node’s address from which it received the RREP. Whenthe RREP reaches the source, the path is established, and the source can send data packets to thedestination, without knowing the exact path to the destination. Information about this RREQ in nodes,which are not part of the established route, times out and will be deleted, because no RREP is received.This mechanism reduces packet overhead, because the complete path needs not to be coded withinevery packet. On the other hand extra memory in every node is needed to save the next hopinformation of all connections using this node.

When an intermediate node notices a movement of a downstream node, it generates a RREP packetwith an infinite next hop metric. The RREP travels upstream to the source, informing all intermediatenodes, about the impossibility of reaching the moved node. All nodes can therefore delete existingroutes, and set up new RREQ messages for their destination.

4.4. Zone Routing Protocol (ZRP)

The Zone Routing Protocol [8, 9] is often called a hybrid ad hoc routing protocol as it combines proactive and reactive elements. The ZRP maintains routing information for a local zone, andestablishes routes on demand for destinations beyond this local neighborhood. It limits the scope of thelocal zone by defining a maximum hop number for the local zone (e.g. 3 hops). Using ZRP with amaximum hop count of zero for the local neighborhood creates a reactive routing algorithm, and usingit with maximum hop count → ∞ creates a pure proactive routing algorithm. A route to a destinationwithin the local zone can be established from the proactively cached tables of the source node. Therouting algorithm used in the local zone can be based on every table-driven routing algorithm, but ithas to be extended in that way, that packets contain the “time to life” (TTL) information, whichdescribes the maximum hop count of the local zone. For routes beyond the local zone, route discoveryhappens reactively. The source node sends a route requests to its border nodes, containing its ownaddress, the destination address and a unique sequence number. Border nodes are nodes which areexactly the maximum number of hops to the defined local zone away from the source. The border

nodes check their local zone for the destination. If the requested node is not a member of this localzone, the node adds its own address to the route request packet and forwards the packet to its border nodes. If the destination is a member of the local zone of the node, it sends a route reply on the reverse

path back to the source. The source node uses the path saved in the route reply packet to send data packets to the destination. To reduce the signaling messages during a route discovery copy-detectionmechanisms are used to detect unnecessary forwarding of route request packets. The main advantageof the ZRP is a reduced number of required RREQ messages and further on the possibility to establishnew routes without the necessity to completely flood of the network. The main disadvantage is theincreased complexity of the routing algorithm, especially the copy-deletion algorithms are difficult toimplement. This also imposes additional requirements to the already limited processing power of mobile nodes.

5. Routing in Peer-to-Peer networks

5.1. The Gnutella Protocol v0.4

The most prominent example for flat routing architectures in the area of Peer-to-Peer networks is thenetwork established by the Gnutella protocol [7]. The Gnutella network is made up of serventsdistributed throughout the world, which are interconnected by TCP/IP connections. Within this virtualoverlay network the servents provide the content and perform routing tasks, to make networking

possible.Every servent is connected dynamically to an average of 7 servents [21], depending on the

bandwidth of the servent’s network connection. The messages, routed via these connections can bedivided into two categories. One type are the query messages, and the second type are the respond



messages. The query messages are used to explore the structure of a terminals neighborhood, bysending out PING() messages. Secondly query messages are used to search for certain content, e.g.mp3 compressed audio files in the network, by sending out QUERY() messages.

The Gnutella network employs a routing concept, known as “viral propagation” for the querymessages. This means that a servent searching for content or exploring the network, sends out a querymessage, i.e. a QUERY() or a PING() message, to all the neighboring servents it is currently directlyconnected to via TCP/IP connections in the virtual overlay network. Thus every servent is able toexplore in a completely decentralized manner without the need for a central entity, by more or lesssimply flooding the network.

The second type of messages, which are used in the Gnutella network, are the respond messages,which are used to answer received query messages. The answer is QUERY_HIT message, if a QUERYwas received, and the peer hosts the demanded content. A PONG-message is used to answer a PINGmessage, and thus to make the querying client aware of its presence. These respond message are of nointerest to the rest of the network, and they therefore have to be routed only to the querying servent. Toavoid flooding, respond messages are routed back to the querying terminal on the same way

backwards, the original query message traveled to the receiving servent.Beside the – application routed - signaling messages, i.e. the respond and query messages, thecontent a servent is querying for must also be distributed through the virtual overlay network.However, to minimize the load on the existing overlay network and especially of its servents/routers,the demanded data is transmitted “out-band”. “Out-band in this context means, that with the address

provided in the QUERY_HIT message, a direct - only IP routed - connection between the querying andthe responding servent is established. This connection is used to transmit the content directly between

both peers.The major problem of the Gnutella protocol v0.4 is that parts of the virtual overlay network are

flooded to a certain extent with ping and query messages, which causes a high signaling load. Toreduce this load, a time-to-live value (TTL) is attached to each query message in the Gnutella protocolv0.4. This means that a query message is only forwarded by a servent, if the TTL-value, which isdecreased with every hop, does not equal to zero. Thus a certain transmission range for these messages

is defined, which prevents the network from being flooded by query messages, which could eventuallylead to scalability problems, especially for servents connected to the network with only a low bandwidth connection.

5.2. Query Routing

Rohrs [20] presented a concept to reduce the number of flooded query messages, by routing thesequeries based on the search keywords. The basic idea of query routing in the virtual overlay network is, that servents exchange their query routing tables with their neighbors periodically. The queryrouting tables contain metadata of hosted content, i.e. keywords, and the corresponding IP-address, of the servent from which the metadata was received. Any incoming query is then analyzed for its searchkeywords, and then compared to the local query-routing table. If one of the search keywords matchesto one or more entries in the routing table, the query is forwarded in the direction, given by the routing

table, instead of being flooded to all neighbors of the servent. If no match with the routing table can befound, the query is forwarded to all neighbors of the servent, as long as the TTL-value of the querymessage has not expired.

To minimize the amount of bandwidth necessary to propagate the routing tables, a variant of BloomFilters [3] are used. This means, that each keyword is hashed, and then all keywords of the content of one servent are compressed in a bitmap. [20]. Thus not a whole set of keywords and IP-addresses hasto be exchanged periodically, but only a comparatively small bitmap. Further on incremental updatescould also be used, if only small changes have taken place since the last routing table has been

propagated to its neighbors.However, the major problem with the implementation of query-routing tables is again, how to keep

them up to date, if the network is very dynamic. The problem is, that routing information for a certainfile A which is hosted by the servent X, may still propagate through the network, although the servent



X is not anymore a member of the network. Thus queries may be directed in a wrong direction, whichleads to useless traffic, and to unsatisfied users, as the content they search for cannot be foundanymore.

A solution to this problem could be to set a timer for every routing table entry. After the expiration,this routing table entry is deleted again, to prevent any misleading routings. Further on, the propagationreach of each routing table must be limited, to prevent the routing table, from propagating through thewhole network. This could be done with a hop-counter, which avoids routing tables from being spreadany further, as soon as a certain value for the hop-count has been reached.

5.3. Dynamic Hierarchical Routing

Due to the eventual scalability problems of flat routing architectures as described in chapter 5.1 and5.2, another P2P routing architecture has recently appeared, which tries to combine the advantages of centralized and decentralized routing. Within this approach, which is the basis for the FastTrack [5]architecture, servents are elected at logon to become so called SuperNodes, if they offer sufficient

bandwidth and processing power. Further on they additionally receive a list of already activeSuperNodes, to which they connect. These SuperNodes thus establish a network, with high bandwidthconnections to each other and high computing power at each SuperNode.

If a servent without sufficient capabilities logs on to this network, it is bound to one of the existingSupernodes in it’s neighborhood. Thus clusters are established, which consist of several “normal”servents, and one SuperNode, which has higher capabilities, i.e. a higher bandwidth connection andhigher computing power. After having connected to a SuperNode, the servent uploads informationabout hosted data on to its SuperNode, and thus becomes “visible” to the overlay network.

As a normal servent has only one connection, i.e. the connection to the SuperNode, it directs queriesonly to the SuperNode of its cluster. Upon receiving a query, the SuperNode first searches its owndatabase, whether the demanded data is available in its own cluster. In this is the case, the IP-addressesof the servents hosting the demanded data is sent back to the querying servent. Otherwise, theSuperNode broadcasts this query in the upper layer, namely the SuperNode layer, within which all

SuperNodes are connected to each other via one or more hops. Every SuperNode, which receives thisquery, searches its database and in case of success sends back the according IP-address of the serventof its cluster, to the querying SuperNode. This SuperNode then forwards this response to the queryingservent. Thus the necessary bandwidth for a successful query can be reduced significantly, as the queryis only broadcasted in a small part of the network, namely the SuperNode layer.

6. Similarities and Differences between mobile ad-hoc networks and Peer-to-Peer networks

6.1. Differences between mobile ad hoc networks and Peer-to-Peer networks

The most important difference between a mobile ad hoc network (MANET) and a peer-to-peer network (P2P) is the motivation to create such a network. A MANET is setup to connect a terminalwith another terminal. Users control the terminals, so the main reason for using a MANET is tocommunicate with other users over the same network. We refer to this as user driven.

The reason to use a P2P network is different. Due to the large capacity of a P2P network, the primary goal of most P2P users is to search for data in the network and do not necessarily want tocommunicate with other users. This data may contain video and music files, exchanged e.g. via thewell known Gnutella network. However using Peer-to-Peer for distributed collaboration on the samedocument via a groupware tool, is gaining more and more attention

In P2P networks, the exchange of data is done with a direct link from one computer to the other.The intermediate computers used to setup a connection between these two computers are neither required nor wanted as relay stations for the transmission of the requested data.



An ad hoc network works different. In a MANET, the route discovery as well as the establishedconnection must use the intermediate nodes. Therefore, data connections are indirect, whereas a P2Pdata connection – from the “Overlay” network view - is a direct link. Figure 4 is a schematic picture,which shows the difference between P2P and ad hoc networks in maintaining an establishedconnection. It should be outlined, that the direct link of a P2P connection is much easier to maintainthan an indirect link over multiple hops of an ad hoc network. Due to the mobility of intermediatenodes the number of route reestablishments increases.

Another difference in the network is the structure of a network in P2P and MANET. A virtuallyoverlay network is created to connect computers in a P2P network. Thus the P2P network is separatedfrom the physical network. The network topology of a P2P network is completely different from the

physical network. Computers directly connected in the P2P overlay network may be separated bythousands of kilometers in the real world. Users of a P2P network are distributed over the whole world.Hence, no estimations about their physical position can be made. In contrast to that, the physical

position of an ad hoc network member can roughly be estimated, because the members are denselydistributed in an area. Beside this, the physical network structure of an ad hoc network can usually be

mapped directly to the logical structure of such a network. Exceptions are subsequently added virtualhierarchies to the logical network structure of the MANET. While a P2P network can possibly spanover the whole planet, the position of a single computer is fixed. In contrast the position of a completeMANET is approximately known, the position of a single node within the MANET cannot beforecasted, due to its mobility.

Figure 3 Schematic comparison of MANET andP2P networks: Difficulty of establishing newconnections over multiple nodes

Figure 4 Schematic comparison of MANET andP2P networks: Difficulty of keeping an establishedconnections alive

There are also some differences in the usage of routing algorithms between P2P networks andMANET. Several publications [24, 25] show, that a proactive ad hoc routing scheme only works in

small MANETs, with a number of nodes less then 100. With larger numbers of nodes, updatemessages must be sent too frequently, so the network is mainly concerned with signaling instead of data traffic. Beside the scaling problem, proactive routing algorithms are feasible in ad hocnetworks, whereas they do not work at all in a P2P network. The reason therefore is, that the load,generated by the usage of a Bellman-Ford [2] routing algorithm, is too high for large P2P networks.Using a proactive routing algorithm in a P2P network, would mean, to transmit lists of subjects of all peers to all other peers, which would certainly fail. Even the transmission of all addresses of themembers of a network is not possible, due to the numerous changes in the availability of singlenodes and the great size of such a network.

As mentioned, the P2P routing algorithms is only executed during a search-query, therefore it can be stated as a pure reactive routing algorithm. Despite the usage of similar reactive routingalgorithms in both networks, the execution differs in both networks. In an ad hoc network the search



ends, i.e. the request is not forwarded anymore, when the destination is found, or an intermediatenode knows an actual route to the destination. In contrast to that, the query request does not stop in a

P2P network, although that servent hosts the searched file. The query request is still forwarded bythe servent and only stops when the TTL field is equal to zero, and the query is the maximumnumber of hops away from the source.

Figure 3 points out another difference between the P2P routing architecture and a reactiveMANET algorithm. The main difficulty for setting up a new connection in P2P networks is togather information for the first hop, because there is no possibility in the P2P-overlay network toreceive information about the structure of the underlying IP network. Whereas in a MANETnetwork, the adjacent nodes within transmission range are the possible nodes for the first hop, whichcan easily be found.

The reliability of the physical channels of the underling network infrastructure differs in P2P andad hoc networks. The reliability of the wired links in a P2P network is very high. During thetransmission of a packet, the probability of an error is extremely low, whereas the probability of alost packet due to overflows in intermediate routers is higher. Because a P2P overlay network is

based on the reliable terminal control protocol (TCP), the P2P network layer does not detect any packet failures.

The properties of wireless links are subject of unpredictable changes, causing numerous biterrors. Therefore, the bit error rate of a wireless link is magnitudes greater then the bit error rate of awired link. Transmissions over a wireless link need an appropriate forward error correction (FEC)and a comprehensive MAC-layer beside a reliable transmission protocol like TCP. Hence the effortto transmit data is higher than in a P2P network.

Thus, as shown in Figure 3 , the creation of a new route over multiple hops is easier in a P2Pnetwork, due to the high reliability of wired links compared to wireless links.

To improve the reliability of a mobile ad hoc network, numerous routing algorithms exist. Mostof them setup multiple backup routes for a single connection. In case the first connection breaks, thesource node can instantly switch to another route, and do not have to establish a new route.However, the processing overhead of creating and maintaining multiple routes is enormous, and the

benefit is not ensured, because more routes will break more often.In contrast to that, in P2P network architectures until now no routing algorithms exist, which may

improve the reliability of the network. This is remarkable, because the reliability of the nodes is aslow as the nodes in a MANET. The reason therefore might be, that firstly, the basic physical datatransmission has still a high reliability. Secondly, the basic idea of a P2P network is the distributedavailability of data, so if one node leaves the network, there still should be enough nodes, whichshare the searched data. Therefore, in fact, in a P2P network is no immediate need for a reliabilityimproved routing algorithm.

The behavior of executing a broadcast is also different at both networks. A P2P network is asingle cast network. It can only generate a virtual broadcast, consisting of a number of single castmessages, one message for every outgoing link. In contrast to that, an ad hoc network always

performs a physical broadcast. Every neighbor within transmission range receives a message broadcasted by a node. On the other hand, a logical single cast message is a physical broadcast, andthe receiver must be determined with help of a logical address within the message. Hence, amessage transmitted from a node as unicast, must be discarded by all nodes unintentionallyreceiving the message.

The fixed position of a P2P servent or the mobile position of an ad hoc node also impact theavailable resources. Resources in this context are transmission bandwidth, battery power, memoryand processing power. While a P2P terminal has nearly unlimited resources, they are certainlylimited in an ad hoc node. Currently, the available resources of an ad hoc node are smaller bymagnitudes compared with the resources of a fixed P2P computer.

Beside numerous analogies of P2P networks and MANET in the field of security, there is oneimportant difference. The usage of a public key infrastructure (PKI) within a P2P network is easier then in an ad hoc network. A computer in the P2P network just needs to know the address of the



PKI to receive a trusted public key of any member in the network, because it can directly connect tothe PKI. On the other hand the knowledge of the address of an PKI in an ad hoc network is not

enough to receive a public key for a secure connection between two nodes. Before receiving thekey, a node must execute a route request to the PKI, not knowing if it can set up a connection to thePKI at all. After the route setup, the node needs a stable route for the transmission of the packetcontaining the public key, before the PKI-connection could be terminated. In fact, the realization of an Internet-like PKI in an ad hoc network is not feasible until know. Therefore, a new structure for aPKI in ad hoc networks needs to be invented.

As a summary of this chapter, it can be stated that wireless ad hoc networks and wired peer-to- peer networks have differences, summed up in Table 1 . However beside all differences there arealso a lot of similarities, as shown in the next section.

Table 1 Differences between Peer-to-Peer and Mobile Ad Hoc Networks Difference P2P network MANET

Reason of creating a network create an abstract/virtualinfrastructure which is more or lessindependent of the physical one

create an initial/physicalinfrastructure for connectivity

Connection between two nodeswired and direct connection (at theP2P layer)

wireless and indirect connection,over several intermediate nodes

Reliability of connections high, due the wired links Low, due to the wireless links

Structurevirtual overlay,

physical structure is separatedform the logical structure

logical and physical network structure correspond

Physical diameter of a network can span the worldmembers are densely distributedin an area, therefore the positionof the network is roughly known

Routing stops when TTL field is 0 stops when destination is found

Proactive routing algorithm not possible with limitations to the network size possible

Reactive routing algorithm possible possible Reliable routing algorithms not required, not implemented exist

Broadcast virtual broadcast, realized withmultiple unicasts physical broadcast performed

Mobility of nodes fixed mobile Available resources: memory,

processor, power supply,transmission bandwidth

practical unlimited limited

Usage of a PKI possible not feasible

6.2. Similarities between mobile ad-hoc networks and Peer-to-Peer networks

The basis of both networks, i.e. Peer-to-Peer and wireless ad hoc networks, is the concept of self organization. In most cases, except the hybrid Peer-to-Peer approach (see [22]), no central entities,which manage and coordinate the network, are given nor is such a network in any form

preconfigured. The network is established, as soon as the single participants decide to create anetwork, by establishing connections to each other. Thus although the nodes stay the same, thenetwork alters permanently, because the nodes most probably change their connections to eachother.

The frequently changing topology is another parallel of Peer-to-Peer and wireless ad hocnetworks and results from the permanent change of connections. In the wireless area, this is caused

by the terminal mobility of the nodes. Thus previously established connections to other nodes may break down, as the node might leave the transmission range of previously neighboring nodes. As aresult new connections have to be established to other nodes which can be reached now.



In Peer-to-Peer networks, the geographical mobility is not an issue, but the frequent on and of switching of the servents itself. The average session duration is only 2 hours per day, which results

in an average availability of 0.083 [21]. Further on the dynamic assignment of IP-addresses byDHCP servers and network address translators causes further problems, as a certain member cannot

be located at a stable IP-address. This could be compared to the geographical mobility of wirelessnodes. As a result, a statement concerning the current network situation can hardly be given, as it isnearly impossible to gather in a short time all the information about the whole network, due to thedynamic network topology.

However, the availability of the terminals is important, as the initial connection request to ademanded objected, may it be a file or a user, and additionally in case of success the possible replymessage must be forwarded via multiple hosts. A direct connection cannot be established, as it isnot known, where a certain object can be found in the network.

Further physical constraints have to be taken into account in the wireless area, as the mobilenodes have a limited transmission range. If one of the intermediate links fails during the abovedescribed setup procedure, it might not be possible, that the query message ever arrives at the

hosting servent, or that the response message can not be routed back to the servent. In the wirelessarea, the intermediate links are even more important, as additionally requested data must betransmitted via these links, as a direct connection as in the Peer-to-Peer networking area, is not

possible.In both networks no hierarchies are given by default. Hierarchies can only be introduced

virtually, with the help of protocols. By introducing for example SuperNodes in Peer-to-Peer networks or cluster heads in wireless ad hoc networks, virtual layers can be added to achieve better scalability. The problem with hierarchies on the other hand is that the network’s topology ischanging frequently, routing tables of a hierarchical architecture might soon be outdated as thenetwork information must be collected permanently in routers of the higher layers. Keeping thisinformation up to date may be more difficult, than in a flat architecture.

A further critical issue which arises from the self organization, is that flooding is necessary to acertain extent in both networks. As both networks are based on a continuously changing topology,the network must be periodically probed, whether certain links and nodes are still available. This isonly possible via broadcasting or flooding messages, as a central management entity is notavailable. Thus the question of up to what number of participants the network is able to scale, is of high interest in the wireless as well as in the Peer-to-Peer area.

To our knowledge, in simulations the number of more than approximately 100 participants inmobile ad hoc networks has not been tested yet. However we assume, that the maximum number will not exceed a few hundred mobile terminals, due to the physical constraints of the mobileterminals, i.e. low bandwidth and low processing power. In the case of Peer-to-Peer, networks withapproximately 60.000 users still operate satisfactory [15]. If more participants with low bandwidthconnections log on, degradations in the performance are expected. The explanation therefore can befound in the high signaling load, which has to be routed by the servents.

A further problem occurring in both networks is, how a new participant can log on to an existingPeer-to-Peer or mobile ad hoc network. As a central managing entity is in general not given, it isnecessary, that any new servent has to find active members of the network. By connecting to theactive nodes, the new node becomes also an active member of the network. To find active membersof the network, a kind of portal is used, which works more or less like a beacon to announce to notyet connected participants, at which addresses active members can be found. Thus any new serventmust know at least the address of the portal, to be able to log onto the ad hoc network. In Peer-to-Peer networks this portal is located under fixed and preconfigured IP-addresses. A new servent firstcontacts the portal, from which it receives a certain number of IP-addresses of active servents, towhich the new servent is now able to connect and thus to become an active member of the Peer-to-Peer network.

In the wireless area a similar procedure is necessary. However in this case not a specific IP-address of a portal must be known, but a specific frequency range, within which the signaling



channel of the wireless ad hoc network is located. On this signaling channel, the new terminal has toannounce its presence, and thus is able to become an active member of the network. However, a

restriction, which does not occur in Peer-to-Peer networks is, that at least one active member of thead hoc network must be located within the transmission range of the new terminal. Otherwise nologon is possible.

Table 2 Similarities between Peer-to-Peer and Mobile Ad Hoc Networks Similarity Peer-to-Peer MANET

Basic routing principle

Virtual broadcast, flooding Physical broadcast, flooding

Network topologyFlat and frequently changing topology,caused by frequent log-on and log-offs

Flat and frequently changing network topology, caused by log-on and log offsand additional terminal mobility of thenodes

Reliability of nodes low low

Connectionestablishment

Hop by hop, via TCP links, whereas thesingle hop path length is not physicallylimited

Hop by hop via radio links, which arethus limited by the radio transmissionrange

Network log on

Via a portal, which is in this case a fixedserver (“beacon server”). The IP addressof the portal must be known for newservents

Via a portal, which is in this case aspecified broadcast radio channel. Thefrequency range of the portal must beknown for new nodes

Scalability

Limited by bandwidth consumingsignaling traffic (flooding) andadditional high user data rates( approx. 50.000-60.000 currently)

Limited by bandwidth consumingsignaling traffic (flooding) andadditional physical constraints( approx. 100 users)

Network management QoS and AAA are difficult to realize, asa central management unit is notimplemented

QoS and AAA are difficult to realizeand additional physical constraints haveto be taken into account

Security

Due to the separation from lower layers,no lower layer security (e.g. IPSEC)useable. Possible solution: end-to-endencryption to establish security in anuntrustworthy environment

No lower layer security concepts for MANET implemented until now.Possible solution: end-to-end encryptionto establish security in an untrustworthyenvironment

Due to the basic principle of self organization and independence of each participant, it is not easyin wireless ad hoc and Peer-to-Peer networks to offer network management functionalities, likeauthorization, authentication and accounting, or even Quality of Service features. Especially the twotopics authorization and authentication lead to the basic question of trust in an untrustworthyenvironment, as a central authority unit is missing. How can e.g. a participant be sure, whether it’scommunication partner, is the one which originally answerd a request?

Concerning secure transmission, no lower layer security approaches, as e.g. IPSEC [1], can beused. The reason is the necessity, that user and signaling data has to be routed at least partly via

other participants of the network. Thus an end-to-end security can not be granted by IPSEC tunnels, but only by encryption techniques implemented on higher layers, e.g. by the use of PGP.As mentioned above, Quality of Service is also an issue which is not easy to solve in any ad hoc

network. Guarantees of connection parameters can hardly be given, as the connections areestablished and managed by independent peers. Thus delay and bandwidth critical applications, likee.g. video communication can hardly be provided in wireless ad hoc or Peer-to-Peer networks,independent from their additional physical constraints.

Although some difficulties arise from the lack of a central authority in Peer-to-Peer and wirelessad hoc networks, such a form of networking also offers the advantage that no additional hardware isnecessary for the establishment and basic management of the network. Communication and dataexchange is made possible in a fast changing environment, where possibly fixed or static network



elements could mean too much overhead. Further on, there is no single point of failure in Peer-to-Peer and ad hoc networks, which makes these networks as whole comparable robust and reliable.

Finally it can be concluded, that there exist similarities between Peer-to-Peer and mobile ad hocnetworks. These similarities are summarized in Table 2 .

7. Conclusion

As outlined above, both networks show similarities, because both networks have a similar goal,namely to establish to communication networks, and additional rely on a similar basis, which is adynamic and frequently changing network topology. Further on, Peer-to-Peer and mobile ad hocnetworks are based on a completely decentralized architecture, without central or preconfiguredentities, than the entry portals. Thus scalability problems arise, as a network architecture has to

provide routing functionalities, as well as the routing information is distributed over all the participants of the network. To solve this scalability problem, dynamical hierarchical concepts have

been introduced in Peer-to-Peer as well as in mobile ad hoc networks.Further on, the decentralization of the network information over all participants is leading to

similar problems concerning the network management. How could guarantees for e.g. a minimum bandwidth and a certain maximum delay be guaranteed, if every node decides on its own whichconnections to establish and maintain, and which to break up?

Although similar tasks and solutions exist in Peer-to-Peer and mobile ad hoc networks, bothnetworks also have fundamental differences (see Table 2 ). They rely on a completely different

physical basis with different constraints concerning bit error rate, bandwidth and other resources.Thus also the services realized until now in ad hoc networks and Peer-to-Peer networks aredifferent. The aim of Peer-to-Peer networks is sharing files of a size of approximately 5 MByte. Incontrast the aim of mobile ad hoc networks is to provide user to user communication and only a dataexchange of small volumes. This fundamental difference affects also the routing goals of bothnetworks. Therefore routing in mobile ad hoc networks could be categorized as destination driven,whereas in Peer-to-Peer networks routing could better be categorized as data driven.

However, similarities between both networks exist (see Table 2 ), and thus possible synergiesshould be used. Problems which come up in one network might already be solved in the other. Theavailable resources in IP based Peer-to-Peer networks are even more capable than in a mobile adhoc network. The routing architecture would have to be modified. Therefore already promisingsolutions in the area of mobile ad hoc networks exist. For example we could imagine to employ azone based routing concept, as described in Section 4.4, in a Peer-to-Peer network. Whatconsequences would arise for the signaling traffic and reliability for a Peer-to-Peer network usingzone based routing? Could possibly a better scalability be achieved? On the other hand, theimplementation of hash keys in routing tables and the usage of Bloom filters in a mobile ad hocnetwork could possibly decrease the signaling traffic and additionally reduce the amount of storagecapacity necessary to store the routing tables.

Especially the separation between the physical and the logical network architecture, as it hasalready taken place in Peer-to-Peer networks, could possibly open up new possibilities in mobile adhoc networks, where the physical and the logical network structure are still the same. A lower network layer (the counterpart to the IP-layer) in a mobile ad hoc network might then be responsiblefor the node to node connections, on which a high layer (the counterpart to the Peer-to-Peer layer)then might establish its virtual connections.

Sometime in the future one could thus imagine a Peer-to-Peer over mobile adhoc network. Due tothe physical constraints given today in mobile networks, there is certainly still a long way to go.However as bandwidths of several 100kbit/s come into discussion along with UMTS and IMT2000,this thought might be seen from another perspective some time in the future. On the other hand,Peer-to-Peer networks might sooner or later not only be used to share files, but also for person-to



person communication. Therefore routing problems certainly still have to be solved, but conceptsalready developed for mobile ad hoc networks might offer a solution to a certain extent.

References

1. R. Atkinson, S. Kent. “Security Architecture for the Internet Protocol”. RFC2401. November 19982. O. Bertsekas, R. Gallager, Data Networks, 2 nd Edition, Prentice Hall Inc., 19923. B. Bloom. “Space/time trade-offs in hash coding with allowable errors”, CACM, 13(7): pp.422-426, July

19704. The Bluetooth Specification, see http://www.bluetooth.com/dev/specifications.asp5. http://www.fasttrack.nu/index_int.html. January 20026. M. Frodigh, P. Johansson, P. Larsson, “Wireless ad hoc networking: The art of networking without a

network” Ericsson Review, No. 4, pp. 248-263, 2000.7. The Gnutella Protocol Specification v0.4. http://dss.clip2.com/GnutellaProtocol04.pdf. August 2000.8. Z. Haas, M. Pearlman, “The Performance of Query Control Schemes for the Zone Routing Protocol”,

IEEE/ACM Transactions on Networking, Vol. 9, No. 4,August 20019. Z. Haas, M. Pearlman, and P. Samar., "Zone Routing Protocol (ZRP)," IETF Internet Draft, draft-ietf-

manet-zrp-04.txt, January 2001.10. D. Johnson, D. Maltz, “Dynamic Source Routing in Ad hoc Wireless Networks”, edited by T. Imielinski

and H. Korth, Kluwer Academic Publisher, pp. 153-181, 199611. D. Johnson, D. Maltz, Y. Hu, "The Dynamic Source Routing Protocol for Mobile Ad Hoc Networks

(DSR)," IETF Internet-Draft, draft-ietf-manet-dsr-06.txt, November 200112. J. Jubin, J. Tornow. „The DARPA Packet Radio Network Protocols”, in Proceedings of the IEEE volume

75, 1, pages 21-32, Jan 198713. A. Langley. “Freenet”. ”. Appeared in “Peer-to-Peer Harnessing the Power of Disruptive Computing”.

Editor: A. Oram. Publisher: O'Reilly & Associates, Incorporated. March 200114. B. Leiner, D. Nielson, “Issues in Packet Radio Network Design, Proceedings of the IEEE Special issue on

“Packet Radio Networks”, 75, 1:6-20, 198715. Limewire. “Current Network Size”. http://www.limewire.com/historical_size.html. January 2002

16.

The Napster Protocol. http://opennap.sourceforge.net/napster.txt. September 200117. C. Perkins, “Highly Dynamic Destination Sequence Distance Vector Routing (DSDV) for MobileComputers”, ACM SIGCOMM'94, 1994

18. C. Perkins, E. Royer, “Ad hoc On demand Distance Vector Routing”, Proceeding of 2 nd IEEE Workshopon Mobile Computing Systems and Applications, February 1999

19. C. Perkins, E. Royer, and S. Das. "Ad Hoc On Demand Distance Vector (AODV) Routing." IETF Internetdraft, draft-ietf-manet-aodv-09.txt, November 2001 (Work in Progress).

20. Christpher Rohrs. “Query Routing for the Gnutella Network”. http://www.limewire.com/developer/query_ routing/keyword%20routing.htm. December 2001.

21. S. Saroiu, P. K. Gummadi, S. D. Gribble. “A Measurement Study of Peer-to-Peer File Sharing Systems”Technical Report # UW-CSE-01-06-02. July 2001

22. R. Schollmeier. “A definition of Peer-to-Peer Networking towards a Delimitation against Classical Client-Server Concepts”. Proceedings of WATM-Eunice 2001. pp. 131-138. 2001.

23. The IEEE 802.11 Standards, see http://grouper.ieee.org/groups/802/11/24. P.Johansson et al. “Scenario Based Performance Analysis of Routing Protocols for Mobile Ad-hoc

Networks”, Mobicom ’99, 1999 Seatle, USA25. J. Broch et al., “A Performance Comaprison of Multi-Hop Wireless Ad hoc Network Routing Protocols”,

MobiCom ’98, 1998, Dallas, USA

Documents

(4)Routing in MANET and P2P Nra Comparison