Mobile Ad Hoc Networks Applications Part 11 doc

MANET Mining: Mining Association Rules 23directly on the number of node MANET size, the distance in terms of number of nodes between the communicating nodes, and the speed of ARM algorit

MANET Mining: Mining Association Rules 19

This swap is equivalent to dropping one of the two similar bit-vectors in the bit-matrix Since

there is utterly no difference between the source and the destination matrices the same MFSs

(key) are obtained

Wormhole attacks do not affect KDTM Wormhole leaks routed packets at a node to the outside world Still the MFSs built from the leaked packets is not the same as that of the end

nodes because not all traffic from the source to the destination pass through the same route

KDTM is immune to Man-in-Middle (MI M) attacks In passive MI M attack, the malicious node just builds and mines its bit-matrix, however, the resultant MFS obtained is different from that of the end nodes Still in this situation, the MFS obtained at the end nodes is not affected because MI M does not alter the bit-vector of both passing data packets and passing ACK Two scenarios are observed in active MI M attacks The first scenario, the

malicious node forges/modifies the bit-vector of the passing data packets This means thesame alteration is reflected in both the bit-matrices of the end nodes In the second scenario,

the MI M alters the bit-vector of the passing ACK This means the same change is induced in

the source bit-matrix but not in the destination bit-matrix The difference induced between

source and destination bit-matrices is insufficient, because a small number of ACK pass through the same route; and therefore, the same MFS obtained at the end nodes.

Notably, active MI M can be identified through checking of bit-vector by routing nodes before

sending it to the next node; and if any node discover that its bit (or the bits of her neighborswho have not received the packet) is changed, then this node should send a warning message

to the other nodes in the MANET that there is an active MI M in the network.

Simulation results show that KDTM is tolerant, in that adding/deleting bit-vectors randomly to/from bit matrix up to 30 % does not change the resultant MFS Further more, KDTM allows concatenating several MFSs or keys in a bid to develop a stronger key.

KDTM may applies Nitin’s watch dog and Pathrater concepts to eliminate malicious

nodes in the transmission range of the end nodes so that the extracted key is notcompromised (Kyasanur & Vaidya, 2003)

KDTM is a new cross layer key distribution scheme, which extracts MFS from network layer

to be used in other layers, for instance, the application layer

6.3 Key revocation

Key disclosure is very frequent in MANET There is no guarantee that the route between the

communicating nodes is free of malicious nodes

In contrast to using static long-term keys, dynamic short-term cryptographic keys can beused to minimize the availability of ciphertext, encrypted with the same key, and therefore,making it difficult to compromise the key (Menezes et al., 1996) Accordingly, key renewal iscompulsory to reduce the amount of disclosed packets in case the key is compromised In thenew method, key renewal, not affected by any other factor and is very simple because the key

is mined as long as there is traffic, may be done at any time

Key can be changed periodically between the two communicating nodes The parameterssuch as Supportσ, Mining Rate Δ and step threshold λ may be changed to mislead the MIM.

This is somehow similar to frequency hopping in wireless communication used for securitypurpose

The next two sections analyze mathematically and experimentally the new framework

341MANET Mining: Mining Association Rules

6.4 Mathematical analysis of the new framework

One of the main features of Apriori algorithm is tolerance, in the sense that arbitrarily addingsome rows (bit-vectors) with random values to the data set (bit-matrix) does not affect the

end result (outcome), and therefore, the same MFS is obtained Further more, deleting some

rows (bit-vectors) randomly from a data set (bit-matrix), does not change the output of thealgorithm At the same time, it is very difficult to guess the output of the algorithm withoutacquiring the whole bit-matrix

The algorithm can be applied on three different types of traffic The first type is the data

traffic The algorithm extracts the MFSs from the bit-matrix of bit-vectors of data packets.

The second type is the acknowledgement traffic and the third type is a mixture of data andacknowledgement packets

Consider a MANET with a set of n nodes The output of Apriori algorithm is MFSs in an increasing order and without repetition The number of ways to form MFS of length i is:

MANET Mining: Mining Association Rules 21

Accordingly, all the possible ways to form an MFS of variable length i is:

C(n, 2) +C(n, 3) + +C(n, i) + +C(n, n−1) +C(n, n)

See figure 7, the sum of the nth row of Pascal triangle is given by (Mott et al., 1992):

C(n, 0) +C(n, 1) +C(n, 2) + .+C(n, i)+ +C(n, n−1) +C(n, n) = 2n (3)From 2 and 3, the total number of ways is:

C(n, 2) + .+C(n, i) + +C(n, n−1) +C(n, n) =2n− (n+1) (4)

If i=2, then the source and the destination are neighbors, that means no intermediate nodes

If i=n then the topology is chained.

Equation 4 assumes that the MFS may contain any number of nodes not exceeding n In fact,

this may be correct in one case only, a chain network topology For example, queue of soldiersfollowing their commander

The number of routing nodes related to several factors, namely the routing protocol,sending/receiving range, and so on

6.5 Experimental analysis of the new framework

In this section, the length of MFSs that are used as tokens (keys), is measured experimentally The NS2 simulator is utilized to generate different scenarios Same parameters that are used

in sections 4 and 5, and listed in table 4, are used in this section except for the density of nodes

In reference to the density of nodes in MANET, Royer (Royer et al., 2001) shows that the

optimum number of neighbors, for 0 m/s mobility or stationary nodes, is around seven oreight per node This number differs only slightly from what Kleinrock proved for a stationarynetwork (Kleinrock & Silvester, 1978) The density of nodes in wireless network is given by:

For example, the average length of the key (MFS) is i=15, which corresponds to the strength

of the key of C(300, 15) =284, using the following parameters for simulation: NMS = 10 m/s;

mining rateΔ=5 s; number of nodes = 300; terrain area = 2700×2700 m2; Supportσ=40 %; routing protocol is DSR; and data traffic.

343MANET Mining: Mining Association Rules

Table 6 The average length of MFS.

6.6 Outstanding features of the new Scheme

Several features make the new scheme more effective, more flexible, more tolerant and more

secure than the present key distribution schemes in MANET These features include:

– Robustness: The protocol is flexible and works in all circumstances, In other words,the absence of any number of nodes in the network topology at any time does notaffect the the new protocol All nodes in other schemes, such as schemes proposed

by (Becker et al., 1998; Burmester & Desmedt, 1994; Kim et al., 2001), should be onlinebefore the key establishment process is completed (Chan, 2004)

– Transparency: The new scheme is transparent and works in all scalable routing protocols.– Packet Size Independence: The new security protocol is independent of the packet size andtype In other words, it operates on all types of traffics, such as data, acknowledgement andcontrol

– Key Revocation and Renewal: The key can be renewed or removed any time even before itsexpiry time These activities reinforce the security of the key

– Overhead at Intermediate Nodes: The new scheme has low overhead on intermediatenodes, achieved through eliminating cryptographical checking of packets at intermediatenodes The present schemes which use public key cryptography have high overhead onintermediate nodes

– Scalability: The new scheme allows the number of nodes to be adjusted Notably, the bigger

the number of nodes in the network the bigger the number of ways to choose MFSs and the

higher the security

– Time and Space Complexities: Experimental results of the new protocol show that the

time-complexity of the protocol for MANETs is of second order These complexities depend

MANET Mining: Mining Association Rules 23

directly on the number of node (MANET size), the distance (in terms of number of nodes) between the communicating nodes, and the speed of ARM algorithms used The space

complexity is Sizeof(bit-vector) * Numberof (bit-vectors), where bit-vectors is equivalent tothe number of contributing packets

– Message Complexity: The new scheme has a message complexity of zero for all routing

protocols For source routing protocols such as DRS , which need not attach the bit-vector

at all because each data packet has its route; still the message complexity is zero Evenfor other protocols the complexity is zero because the bit-vector is attached to packets, andtherefore, no security-dedicated packets are sent

– Fault Tolerance: The failure of a number of nodes does not affect the new protocol becausethe same bit-entries are dropped from all bit-vectors

– Adjustability: The new scheme is adjustable For instance, Apriori is tunable through

the Support parameter of MFS, size of bit-matrix and bit-vector extraction time It is not

necessary to attache bit-vector to each packet

7 Conclusion and future research directions

KDTM, a cross layer scheme, shows that MANET traffic in the third layer is raw material

that can be mined and utilized in other layers In addition, the scheme shows how to collect

dynamic data from complex and chaotic MANET with large population of mobile nodes and convert it into knowledge The algorithm mines the MFS patterns through ARM technique employing two methods TAR and SAR mining.

The new concepts generated by KDTM and this chapter as a whole can be extended in several

ways Described below are some of the possible enhancements and extensions:

– Security Enhancement: MANET mining techniques can be used in identifying malfunctioning or blackholes or compromised nodes in MANETs through analyzing the MFSs Such nodes, if identified by a number of other nodes in MANET, are

discarded/excluded from the list of trusted nodes

– Maximizing the Network Life Span: Energy conservation is of paramount importance

in MANET, therefore, uniform energy consumption of nodes increases considerably the lifetime of the network MFS can be used to identify active and dormant nodes Dormant nodes in MANET increase the workload on active nodes and thereby decreasing their

lifespan It is therefore evident that decreasing the number of dormant nodes translates

into increasing the life span of the MANET Accordingly, MFSs may be considered as a life

span metric

– Load Balancing: Heavily-loaded nodes may become a bottleneck that lowers the network

performances through congestion and longer time delays MFSs can be used as an indicator

to avoid over utilized nodes and select energy rich nodes for routing

– Activity Based Clustering: Similar to other clustering metrics, like power, distance andmobility, among others, node activity levels can be considered as a metric for cluster

formation Nodes belonging to one MFS (pattern) are most likely connected and can be

used as a cluster Another metric for clustering is the Support parameter, i.e., the higher theSupport level the higher the relationship among the routing nodes

– Routing and Multicasting: Nodes belonging to one MFS are most likely connected Accordingly, delivery or sending of packets is guaranteed amongst nodes in the same MFS.

345MANET Mining: Mining Association Rules

– Applying Different Association Rules Mining Types: This chapter applies positiveassociation rules mining techniques that mine binary attributes and considers that theutilities of the itemsets are equal The frequency of an itemset may not be a sufficientindicator of interest Non-boolean fuzzy association rule mining such as weighted/utilityassociation rules, may find and measure all the itemsets whose utility values are beyond

a user specified threshold that suggest different decisions For example, in battlefield acommander can give higher weight/utility to his higher rank commanders and less weight

to soldiers in order to find the hidden relationships (rules) amongst them These rules maygive an idea about soldiers who are in touch with each other, with commanders, and so on

– Wireless Sensor Networks (WSN) has the inherent characteristics of MANETs, and therefore, the aforementioned benefits of using MFS in MANETs may also be applicable

in WSN.

8 References

Agrawal, R., Imielinski, T & Swami, A (1993) Mining association rules between sets of items

in large databases, Proceeding of the 1993 ACM SIGMOD International Conference on Management of Data, ACM, New York, NY, USA, Washington, D.C., United States,

pp 207–216

Agrawal, R & Shafer, J C (1996) Parallel mining of association rules, IEEE Transactions on

Knowledge and Data Engineering 8(6): 962–969.

Asuncion, A & Newman, D J (2007) UCI machine learning repository

URL: http://www.ics.uci.edu/∼mlearn/MLRepository.html

Becker, K., Wille, U & Wille, U (1998) Communication complexity of group key distribution,

Proceedings of the 5th ACM conference on Computer and communications security, ACM

New York, NY, USA, San Francisco, California, United States, pp 1–6

Burmester, M & Desmedt, Y (1994) Vol 950/1995 of Lecture Notes in Computer Science,

Springer Berlin, Heidelberg, chapter A Secure and Efficient Conference KeyDistribution System, p 275

Chan, A C F (2004) Distributed symmetric key management for mobile ad hoc networks,

Proceeding of IEEE INFOCOM 2004 Twenty-third Annual Joint Conference of the IEEE Computer and Communications Societies, Vol 4, IEEE Press Piscataway, NJ, USA, Hong

Kong, pp 2414–2424

Fall, K (2007) The NS Manual, The VINT Project, University of California.

Fard, A M & Ester, M (2009) Collaborative mining in multiple social networks data

for criminal group discovery, International Conference on Computational Science and Engineering, IEEE CS Digital Library, Vancouver, Canada, pp 582–587.

Frawley, W J., Piatetsky, G & Matheus, C J (1992) Knowledge discovery in databases: An

overview, AI Magazine 13(3): 57–70.

Fumy, W & Landrock, P (1993) Principles of key management, IEEE Journal on Selected Areas

in Communications 11(5): 785–793.

Ghoreishi, S M & Analoui, M (2009) Design a secure composite key-management scheme

in ad-hoc networks using localization, International Journal of Computer Science and Network Security 9(9): 35–49.

http://www.isi.edu/nsnam/ns/tutorial/

Hegland, M (2005) Wspc/lecture notes series: The apriori algorithm - tutorial, Technical

report, Australian National University, CMA, John Dedman Building, Canberra ACT

MANET Mining: Mining Association Rules 25

0200, Australia

Hofmann, M (2003) The development of a generic data mining life cycle (dmlc), Master’s thesis,

MSc in Computing Science , Dublin Institute of Technology, Duplin, USA

Jabas, A., Abdulal, W & Ramachandram, S (2010) An efficient and high scalable key

distribution scheme for mobile ad hoc network through mining traffic meta-data

patterns, Fifth IEEE International Conference on Network and System Security (IEEE NSS’10), IEEE CS Digital Library, Melbourne, Australia.

Jabas, A., Garimella, R M & Ramachandram, S (2008a) Manet mining: Mining

step association rules, Fifth IEEE International Conference on Mobile Ad-hoc and Sensor Systems (IEEE MASS’08), IEEE CS Digital Library, Atlanta, Goergia, USA,

pp 589–594

Jabas, A., Garimella, R M & Ramachandram, S (2008b) Manet mining: Mining temporal

association rules, Third International Workshop on Intelligent Systems Techniques for Ad hoc and Wireless Sensor Networks (IEEE IST-AWSN 2008), Sydney, Australia, IEEE CS

Digital Library, Sydney, Australia, pp 765–770

Jabas, A., Garimella, R M & Ramachandram, S (2008c) Proposing an enhanced mobile ad

hoc network framework to the open source simulator ns2, Mosharaka International Conferences on Communications, Computers and Applications (IEEE MIC-CCA’08), IEEE

CS Digital Library, Amman, Jordan, pp 14–19

Javaheri, S H (2007) Response modeling in direct marketing, a data mining based approach for

target selection, Master’s thesis, Continuation Courses, Marketing and e-commerce,

Department of Business Administration and Social Sciences, Division of Industrialmarketing and e-commerce

Joe, B (2009) Do association rules represent supervised or unsupervised learning, Technical

report http://wardselitelimo.com/2009/07/02/.

Kim, Y., Perrig, A., & Tsudik, G (2001) Communication-efficient group key agreement,

In 17th International Information Security Conference (IFIP SEC01), Kluwer Academic

Publishers Norwell, MA, USA, Paris, France, pp 229–244

Kleinrock, L & Silvester, J (1978) Optimum transmission radii for packet radio networks

or why six is a magic number, Proceedings of the IEEE National Telecommunications Conference, IEEE CS Digital Library, Birmingham, Alabama, p 4.3.14.3.5.

Kyasanur, P & Vaidya, N H (2003) Detection and handling of mac layer misbehavior

in wireless networks, International Conference on Dependable Systems and Networks (DSN’03), IEEE CS Digital Library, San Francisco, California, pp 173–182.

Lamport, L (1987) Synchronizing time servers, Technical report, Digital Equipment

Corporation Systems Research Center

Luo, H., Kong, J., Zerfos, P., Lu, S & Zhang, L (2003) Ursa: Ubiquitous and

robust access control for mobile ad-hoc networks, 58th IEEE Vehiclular Technology Conference VTC’03, Vol 3, IEEE Press Piscataway, NJ, USA, Orlando, Florida, USA,

pp 2137–2141

Menezes, A., Oorschoot, P V & Vanstone, S (1996) Handbook of Applied Cryptography, CRC

Press, San Antonio, Texas

Mott, J L., Kandel, A & Baker, T P (1992) Discrete Mathematics for Computer Scientists and

Mathematicians, Reston Publishing Company, Inc.

ns2 (2009) The network simulator (ns2), Information Sciences Institute

URL: http://nsnam.isi.edu/nsnam/index.php/Main-Page

Olson, D L & Delen, D (2008) Advanced Data Mining Techniques, Springer, Verlag Berlin

347MANET Mining: Mining Association Rules

Post, G V (2005) Database Management Systems: Designing And Building Business Applications,

McGraw-Hill, Irwin

Pujari, A K (2001) Data Mining Techniques, Universities Press, 3-6-747/1/A and 3-6-754/1,

Himayatnagar, Hyderabad 500 029, Andhra Pradesh, India

Rashmi (2009) Manet (mobile adhoc network),

http://www.saching.com/Article/MANET Mobile-Adhoc-NETwork–/334 [Access time: 20 Oct., 2009]

Robinson, J A (2007) Connecting the edge: Mobile ad-hoc networks (manets) for network

centric warfare, Technical report, AIR UNIV MAXWELL AFB, Maxwell-Gunter Air

Force Base Montgomery, Alabama, USA

Royer, E M., Melliar-Smith, P M & Mosery, L E (2001) An analysis of the optimum node

density for ad hoc mobile networks, IEEE International Conference on Communications, ICC, Vol 3, IEEE CS Digital Library, Helsinki, Finland, pp 857–861.

Santoro, N (2007) Design and Analysis of Distributed Algorithms, John and Wiley and Sons, Inc.

Hoboken, New Jersey, Hoboken, New Jersey

Simons, B., Welch, J L & Lynch, N (2006) Fault-tolerant distributed computing, Vol 448/1990 of

Lecture Notes in Computer Science, Springer, Berlin / Heidelberg, chapter An overview

of clock synchronization, pp 84–96

Simovici, D A & Djeraba, C (2008) Mathematical Tools for Data Mining, Set Theory, Partial

Orders, Combinatorics, Springer-Verlag Limited, Uk, London.

Tan, P.-N., Steinbach, M & Kumar, V (2006) Introduction to Data Mining, Addison-Wesley.

Yao, J., Li, X & Jia, L (2003) A new method based on ltb algorithm to mine frequent itemsets,

International Conference on Machine Learning and Cybernetics, IEEE CS Digital Library,

Xian, China, pp 71–75

Yi, S & Kravets, R (2003) Moca: Mobile certificate authority for wireless ad hoc netwroks,

Proc of the 2nd Annual PKI Research Workshop (PKI), National Institute of Standards

and Technology, Gaithersburg, USA

Yi, S & Kravets, R (2004) Composite key management for ad hoc networks, The First Annual

International Conference on Mobile and Ubiquitous Systems: Networking and Services MobiQuitous’04, IEEE CS Digital, Boston, USA, pp 52–61.

0 Wired/Wireless Compound Networking

Juan Antonio Cordero1, Emmanuel Baccelli1, Philippe Jacquet1and Thomas Clausen2

networks are a priori unpredictable and may change dynamically during the lifetime of the

network, no assumptions can be made in general concerning topology, link reliability, routerspositions, capabilities, and other such aspects Routing protocols operating within an AS

– i.e interior gateway protocols (IGP) – must enable each router to acquire and maintain

the information necessary to forward packets towards an arbitrary destination in the routingdomain Currently, the dominant IGP technology is link state routing, as acknowledged byreports of Cisco Systems, Inc such as Halabi (2000)

Routing protocols that were designed for wired, static environments do not perform well

in ad hoc networks: even for small networks, as Henderson et al (2003) points out, control

traffic explodes in a wireless, dynamic context Many efforts have been deployed over the lastdecade, aiming at providing routing protocols suitable for ad hoc networks In such context,information acquisition and maintenance has to be provided by distributed mechanisms,since neither hierarchy nor centralized authority can be assumed to exist Moreover, thetypical bandwidth scarcity experienced in wireless ad hoc networks calls for mechanismsthat are extremely efficient in terms of communication channel utilization In the realm oflink-state routing two main strategies have been explored: (i) the design of ad hoc specificrouting protocols; and (ii) the reuse and adaptation of existing generic routing protocols sothat they can handle ad hoc conditions The first strategy has mainly led to the emergence

of the Optimized Link State Routing protocol, OLSR, standardized as RFC 3626 (2003) Thesecond approach has led to protocol extensions such as RFC 5449 (2009), which enable theoperation of Open Shortest Path First (OSPF) on ad hoc networks

This chapter focuses on scenarios where the AS consists in compound networks: networks

gathering both potentially mobile ad hoc routers, and fixed wired routers Such scenariosmay become frequent in a near future where wireless ad hoc and sensor networks play anincreasing role in pervasive computing Obviously, it is possible to employ multiple routing

protocols within a compound network (e.g one for wireless ad hoc parts of the network,

and another for the wired parts of the network) However, a single routing protocol makesmore economical sense for the industry, and furthermore avoids the potential sub-optimality

of having to route through mandatory gateways between different routing domains Thus asingle protocol is desired to route in compound networks, and (ii) is deemed the best strategy

16

to do so The main reason for this is, that (ii) takes advantage of wide-spread, generic protocolswhich on one hand already provide very elaborate modules for various categories of wirednetworks, and on the other hand can easily accommodate a new module for efficient operation

on ad hoc networks

This chapter thus explores techniques that enable efficient link state routing on compoundnetworks These techniques rely on the selection and maintenance of a subset of links in

the network (i.e an overlay) along which the different operations of link-state routing can

be performed more efficiently The following provides a formal analysis of such techniques, aqualitative evaluation of their specific properties and example applications of such techniqueswith a standard routing protocol

1.1 Terminology

In this chapter, the following notation is used:

– The 1-hop and 2-hop (bidirectional) neighborhoods of a router x are denoted by N(x)and

N2(x), respectively

– The usual notation of graph theory is assumed: G= (V, E)stands for a (connected) network

graph, in which the set of vertices is V=V(G)and the set of edges is E=E(G) Overlay

subgraphs are denoted accordingly, as subsets of G.

– Given two vertices (routers) x, y ∈ V, dist(x, y)is the cost of the optimal path between x and

y Similarly, given two vertices x, y ∈ V reachable in 2 hops, it will be denoted by dist2(x, y)

the cost of the optimal path between x and y in 2 hops or less (local shortest path) For two neighbors x and y, m(x, y) =m(xy)denotes the cost of the direct link from x to y.

1.2 Chapter outline

The chapter is organized as follows Section 2describes the key operations providing link-staterouting Section 3 elaborates on the constraints that ad hoc networking imposes on link-staterouting, with a specific focus on compound networks Section 4 introduces to the notion ofoverlay for performing these key operations, analyzes the properties of several overlay-basedtechniques and discusses their advantages and drawbacks of their use in the context of aconcrete routing protocol Section 5 applies and evaluates the performance of such techniques

as ad hoc OSPF extensions Finally, section 6 concludes this chapter

2 Communication aspects in link-state routing

This section provides a structural high-level description of the operations of link-state routing.Section 2.1 presents a short summary of link-state routing Sections 2.2, 2.3 and 2.4 describe

in more detail the main tasks associated to such operation: neighbor discovery, networktopology dissemination and route selection for data traffic, respectively

2.1 Link-state routing overview

Link-state routing requires that every router learns and maintains a view of the networktopology that is sufficiently accurate to compute valid routes to every possible destination.This, typically (as for OSPF or IS-IS1), in form of shortest paths w.r.t the metrics used.Such shortest paths are computed among the available (advertised) set of links by means

Wired/Wireless Compound Networking 3

of well-known algorithms such as Dijkstra (1959), and will provide effectively optimal routeswhen the view of the topology is up to date

These objectives require that every router in the network performs two operations, other thanthe shortest path computation: first, take efficient flooding decisions for the forwarding oftopology information messages; and second, describe accurately its links in order to advertisethem to the rest of the network Three tasks emerge thus as necessary for the performance oflink-state routing operation:

1 participation in the flooding of topology information (both of self-originated messages and

of messages from other routers),

2 selection of links to advertise to enable shortest route construction and,

3 discovery and maintenance of the neighborhood, as a pre-requisite for the two previoustasks

2.2 Neighbor discovery and maintenance

The discovery and maintenance of neighbors is a prerequisite for performing efficientlink-state routing Without neighborhood knowledge, link-state routing can only be deployed

by means of pure flooding, which has been proven by Ni et al (1999) to be dramatically inefficient when dealing with ad hoc networks (the broadcast storm problem); or with

counter-based or similar approaches, which have severe performance limitations, as shown

in Tseng et al (2003) The most widespread and basic mechanism for neighbor sensing

consists of the periodic transmission of Hello packets by every router in the network (Helloprotocol) Exchange of such Hello packets enable routers to learn their neighborhoods andestablish bidirectional communication, if possible, with neighbors within its coverage range.Aside from this use, Hello exchange may be useful for acquiring additional informationabout the neighbors (geographic position, remaining battery power, willingness to acceptresponsibilities in communication), the links to them (link quality measures) or the neighbors

of such neighbors (2-hop neighborhood acquisition)

2.3 Topology information dissemination

Consistency of the distributed LSDB and correctness of routing decisions require that everyrouter maintains an updated view of the network topology When a router detects a relevantchange in its neighborhood, it needs to advertise it by flooding a topology update message,

so that any other router can modify accordingly its link-state database and, if necessary,recalculate optimal routes

In ideal conditions2, such mechanism would be sufficient for keeping identical LSDBs inevery router in the network Since these conditions are not found in wireless ad hoc scenarios,additional mechanisms might be considered:

– Reliable flooding of topology messages Reception of such messages is acknowledged

by the receiver, or retransmitted by the sender/forwarder in the absence of suchacknowledgment, in a hop by hop fashion Reliable flooding is provided by the main wiredrouting protocols (OSPF, IS-IS), but its cost in mobile ad hoc networks discourages its use

in MANET-specific solutions such as OLSR

– Periodic re-flooding of messages. After a certain interval, even if no changes havebeen registered in the neighborhood, the routers reflood to the network an advertisement

351Wired/Wireless Compound Networking

containing the current state of the links between themselves and their neighbors The length

of the interval is typically related to the mobility pattern of the network: the faster nodes

in the network move, the shorter the interval between consecutive topology messages fromthe same source needs to be

– Point-to-point link-state database synchronization A link between two routers is said

to be synchronized when the routers have completed a synchronization process of their

respective LSDB This involves the exchange of the database contents and the installation ofthe most updated topology information in each of them This mechanism is implemented

in the major wired routing protocols (OSPF, IS-IS), but the conditions in which suchsynchronization is performed are not completely adapted to mobile ad hoc operation.Therefore, the mechanism as-is is not considered in specific protocols such as OLSR, andits use is widely restricted, for instance, in the different OSPF MANET extensions

These mechanisms handle different issues concerning topology dissemination Reliabletransmission permits overcoming phenomena such as wireless channel failures or collisions.Periodic re-flooding and point-to-point synchronization provide up-to-date topologyinformation to routers appearing in the network after some of the disseminated messageswere flooded across the network Periodic reflooding by itself enables every router toacquire the latest topology information (maybe with a non-negligible delay, depending onthe re-flooding interval) In contrast, full synchronization is not capable on its own to assuredatabase convergence from all routers in link-state routing3 Point-to-point synchronization is,

at best, a complementary mechanism to periodic re-flooding that allows a router that has notreceived all the topology updates to get within a shorter delay the last topology informationfrom an updated neighbor

Synchronization techniques implicitly introduce the concept of a synchronized overlay A router

is included into the synchronized overlay if it is aware of the last topology update messages that were flooded across the network, and, correspondingly, it is removed from the overlay

when it does not receive one of more topology information messages In that context, theperiodic re-flooding of topology messages permits including every reachable router into thenetwork within a maximum delay equal to the interval between two consecutive refloods.Point-to-point LSDB synchronization between a router and a synchronized neighbor permits,

in turn, including routers immediately into the overlay (by means of the database exchange

process), i.e., to restore or establish for the first time the router’s synchronism with the rest of

the network

In wired networks, the synchronized overlay is expected to grow monotonically until itcontains all routers – then the network is said to converge Router removals from thesynchronized overlay are rare events mostly caused by physical link disconnections or routershut-downs In ad hoc networks, the nature of the synchronized overlay is far more unstable.Alternative inclusion and removal events may thus occur due to router mobility or wirelesslink quality variations, preventing the network to converge in the usual sense

2.4 Route selection for directed communication

The final goal of any routing protocol is that every router is able to route traffic to any otherrouter (and any destination provided by such router) in the network For a link-state routing

to converge through repeated database synchronization processes In the considered link-state context, synchronization occurs once in a link lifetime, which is not sufficient for assuring convergence.

Wired/Wireless Compound Networking 5

protocol, such ability is provided by disseminating the topology updates of all routers acrossthe network Such dissemination permits every router to construct and maintain updatedrouting tables, as Figure 1 describes schematically

Shortest Path Tree

Fig 1 Construction of the routing table for a link-state routing protocol

The tree of the optimal routes to every destination (Shortest Path Tree) is then computed

by means of well-known minimum paths algorithms Typically, link-state routing protocols(OSPF, IS-IS, OLSR) use Dijkstra (1959), while distance-vector protocols (RIP4, EIGRP5) rely

on Bellman-Ford [Bellman (1958); Ford & Fulkerson (1962)] These algorithms operate over agraph in which vertices correspond to routers in the network and edges mostly correspond

to links advertised by the received topology update messages6 The routing table is thusextracted from the next hop, according to the Shortest Path Tree, to every possible destination

In general, the reconstructed link-state database should bring every router exactly the sameperspective of the network topology, which would require that all links are advertised Inpractice, the set of links that a router advertises to the rest of the network can be restricted asfar as it does not prevent the shortest path algorithm to select network-wise optimal routes

3 Link-state routing with ad hoc constraints

This section exposes the main challenges for link-state routing in ad hoc networks These aremainly related to (i) the efficient dissemination of topology information across the network,

in presence of lossy channels and dynamic topologies as is typical in these networks, and (ii)the ability of the network to acknowledge and react quickly to topology changes Section 3.1presents the most relevant implications of the ad hoc nature in the performance of link-staterouting, while section 3.2 focuses on the specific case of compound networks integrated bywired and wireless groups of routers

3.1 General issues of ad hoc link-state routing

Wireless ad hoc networking presents a certain number of unique communication conditionsthat link-state routing needs to accommodate:

– Unreliability of wireless links Wireless links are inherently unreliable: channel failures

and collisions are more frequent than in wired links Wireless link quality can be alsohighly dynamic Both circumstances make necessary continuous monitoring of the stateand characteristics of links

RFC 2080 (RIPng, designed for IPv6).

previous IGRP.

explores some techniques in which some additional edges, not advertised in such messages, might be included as well.

353Wired/Wireless Compound Networking

– Semibroadcast nature of wireless multi-hop communication Wireless communication

entails shared bandwidth among not only the routers participating in the communication,but also those within the radio range of the transmitting routers This reduces drasticallythe available bandwidth for a router, since it is affected by the channel utilization of itsneighbors Applications may take advantage of such bandwidth sharing phenomenon byprivileging, when possible, multicast transmissions in place of a unicast (point-to-point)approach that no longer corresponds to the physical conditions of communication

– Asymmetry and non-transitivity of links Semibroadcast communication also implies that

the set of nodes receiving a transmission if not (necessarily) the whole network Moreover,the set of nodes receiving a transmission may be different for two routers, even when suchrouters are neighbors This means that wireless links in a multi-hop ad hoc network cannot

be expected to be transitive: the fact that a router x can directly communicate with routers

y and z does not imply that routers y and z can also communicate directly (x ↔ y, y ↔ z

x ↔ z) Asymmetric links (i.e., links in which a router can hear the other’s transmissions, but

not the other way around) are also possible due to specific channel conditions or differentrouter capabilities

– Topology acquisition and maintenance. Neither hierarchy nor specific routers

relationships can be a priori assumed in an ad hoc network Dynamic configuration of

hierarchical schemes becomes unfeasible due to difficulties on electing top-level routers(related to non-transitivity of links) and cost of performing hierarchy recompositions(caused by node failures, node mobility or channel quality variations) Distributedapproaches are thus encouraged in place of hierarchical ones Moreover, unreliability ofwireless links makes necessary to complement topology dissemination with a periodic andfrequent reflooding of topology messages that ensures that nodes acquires the last updateswith a relatively short delay

3.2 Dissemination in compound networks

In addition to wireless ad hoc routers, compound networks also contain wired staticcomponents, for which the typical link lifetime is much higher than for standard ad hoccommunications The coexistence of wired and wireless ad hoc components poses someadditional constraints to those presented in the previous section 3.1 Frequent floodingupdates from the wired components lead to inefficient use of the available bandwidth, asthe information about wired links carried by consecutive messages would be unchanged.Low update frequencies (with intervals in the order of wired networks) may however

be insufficient to accommodate communication failures in the wireless and/or mobilecomponents of the network

Link synchronization between selected pairs of neighboring routers (in addition to topologychanges flooding and periodic topology reflooding) helps to alleviate this issue Point-to-pointlink synchronization enables highly dynamic routers to acquire updated topology informationfrom wired links even long time after its origination, without requiring frequent refloods ofthe same link-state description by the corresponding wired (stable) source

Consider Figure 2, where fixed routers (1 and 2) can handle changes in their wired (stable)links by transmitting topology updates at relatively low rate (with the time interval betweenupdates in the order of minutes) Mobile routers (such as 5, 6 and 7) and, more in general,routers maintaining wireless links (also the hybrid routers 3 and 4) should use significantlylower time intervals (in the order of seconds, depending on their mobility pattern) If, for anyreason, a mobile router (such as 5, 6 or 7) did not receive a topology update from a wired one

Wired/Wireless Compound Networking 7

as router 1, it will be unable to update its LSDB until the next flooding from the wired router,failing at computing valid routes that involve that router in the meanwhile

Legend

Fixed node Mobile node Wired interfaces Wless interfaces Wired/wless ifaces Wired link Wireless link

Fig 2 Example of compound (wired/wireless) network

The inclusion of a LSDB synchronization mechanism addresses the coexistence of wired andwireless components without having to reflood unnecessary topology updates from wiredrouters nor compromising the accuracy of network topology view of ad hoc (mobile) routers.This, at the expense of an additional dissemination mechanism (in addition to regular flooding

of topology changes and periodic topology reflooding) and the corresponding additionalcomplexity in the flooding operation

4 Overlay techniques for compound networks

This section proposes and analyzes various techniques for performing link-state routing in adhoc compound networks Section 4.1 introduces the notion of overlay and reformulates themain operations of link-state routing in terms of overlays Subsequent sections 4.2, 4.3 and 4.4describe three overlay-based techniques (Multi-Point Relays, Synchronized Link Overlay andSmart Peering, respectively) and analyze their most relevant properties, both from theoreticaland experimental (simulation-based) perspectives

4.1 The notion of overlay

The three main operations of link-state routing in ad hoc networks can be reduced to overlay

definition problems Intuitively, an overlay of an ad hoc network is a restricted subset of routers

and links of the network in which a certain operation is performed More formally, the overlay

of a network graph G= (V, E)corresponds to a subgraph S ⊆ G containing a subset of vertices

V(S ) ⊆ V(G) =V and a subset of links E(S ) ⊆ E(G) =E of the underlying network graph G.

In an ad hoc network, link-state routing operations are performed locally (independently byevery router in the network) and thus, the corresponding overlays are built in a distributedfashion and may change dynamically during the network lifetime Three different types ofoverlays can be identified, one for each of the following operations:

– Topology update flooding The flooding overlay has to be dense (in the mathematical

sense) in every of its connected components – meaning that, in case the overlay is not

connected, each of its pieces is at distance ≤1 (number of hops) of every router in thenetwork This condition guarantees that a topology update generated in any of suchcomponents reaches all routers Due to the impact of any additional router in the floodingoverlay (an additional transmission, and the corresponding utilization of the channel of allits neighbors for every topology update generated in the network), the size of such overlayshould be minimized

– Point-to-point synchronization The synchronized overlay contains links between those

routers having exchanged their LSDBs Formally, such overlay needs to form a spanning

355Wired/Wireless Compound Networking

connected subgraph of the general network graph7, in order to facilitate the distribution

of the LSDB over the whole network The number of LSDB synchronization processesinduced by a synchronized overlay is related to the overlay density (the number oflinks in the overlay), and also depends on the lifetime of the synchronized links (giventhat synchronization is performed once during the existence of the link) Therefore,minimization of overhead caused by LSDB synchronization requires a low density overlaywith stable links

– Topology selection In wired deployments, all links are typically advertised to ensure that

all routers in the network have an identical view of the network topology In wireless adhoc networks, this condition is often relaxed, and every router is only expected to acquire

a consistent topological view of the network accurate enough to perform correct routecomputation Hence, selection of advertised links trades-off the size of the topology updatemessages and the accuracy of the topological view of the network in all routers A topologyselection rule must, however, produce a connected and spanning subgraph (otherwise therewould be non-reachable destinations) and whose set of edges contains all network-wideshortest paths – otherwise the computation would be asymptotically suboptimal8

Table 1 summarizes the requirements of each operation to the corresponding overlay

Table 1 Summary of overlay requirements

4.2 Multi-point relays – MPR

Multi-Point Relaying (MPR) is primarily a technique for efficient flooding It reduces thenumber of required transmissions for flooding a message to every 2-hop neighbor of the

source by allowing a restricted subset of 1-hop neighbors (multi-point relays of the source) to

forward it Figure 3 illustrates that a clever election of 1-hop neighbors as relays can achievethe same coverage as allowing every 1-hop neighbor to transmit (pure flooding, see Fig 3.a)while reducing significantly the number of redundant transmissions

The subset of selected relays must satisfy the condition of full 2-hop coverage:

MPR coverage criterion Every 2-hop neighbor of the computing router must be reachable by

(at least) one of the selected multi-point relays

Therefore, an MPR set of a router x can be formally defined as follows:

7I.e., has to include every vertex (router) in the network.

transmission failures and such Asymptotic suboptimality implies that even in ideal conditions (message

would be suboptimal.

Wired/Wireless Compound Networking 9

Fig 3 (a) Pure flooding vs (b) flooding based on the Multi-Point Relays (MPR) principle.

Solid dots in (b) represent multi-point relays.

Different heuristics can be used for selecting multi-point relays, all valid as long as theysatisfy the MPR coverage criterion This chapter uses the heuristic in Figure 4, presented

and analyzed in Qayyum et al (2002).

MPR(x ) ←− { y excl ∈ N(x): y excl provides exclusive coverage to one or more 2-hop neighbor(s) of x }

while (∃ uncovered 2-hop neighbors of x),

Fig 4 Summary of the MPR heuristic

This heuristic assumes that the source is aware of its 2-hop neighbors Acquisition of the 2-hopneighborhood is thus required Dependence on 2-hop neighbors has yet another side effect onthe MPR properties: given that an MPR selection may become obsolete due to a change in the2-hop neighborhood of the computing source, stability of the MPR set is not only affected byconditions in MPR links9, but also by the MPR recalculations due to changes within the 2-hopneighbors or they way in which they are connected to the 1-hop neighbors of the source (seeFigure 5) Such sensitiveness of the MPR set of a router to variations in its 2-hop neighborhoodhas further implications for the MPR overlay that will be further detailed in section 5

S 1 2 3

4 5 6 7

S 1 2 3

4 5

6 7

6

Fig 5 MPR recalculation due to changes in the 2-hop neighborhood Solid dots represent relays

of router S.

4.2.1 MPR as a flooding overlay principle

MPR flooding introduces a directed overlay for every flooded message, by allowing a router

to forward such message if and only if the following two conditions are satisfied:

357Wired/Wireless Compound Networking