Network Mobility and Fragmentation in Wireless Mesh

requirements of the Internet of things.The Internet of things is going to push more devices

into the fringe Internet. It is the diversity and fluidityof the fringe Internet topology that will show the in-adequacies of current routing protocols that follow thehistorical Internet model. The Internet of things willresult in fluid autonomous sections that will roam thefringe Internet requiring temporal partnerships and cooperation to provide the required connectivity. Almostanalogous to the core Internet, these autonomous regionswill need to foster relationships with other autonomouszones providing customer provider, peering, mutual tran-sit and mutual backup relationships but on a much moremicro dynamic scale.

This paper investigates the relevant challenges thatrelate to IP mobility standards and the routing protocolsthat support the formation of mobile mesh networks. Theremainder of this paper is organised as follows. Section2 reviews the current node mobility standards. Section 3details the protocols that maintain mobile network seg-ments. Section 4 details the research challenges. Section5 concludes the paper.

II. NODE MOBILITY

Node mobility permits a single device to leave itscurrent point of attachment and join a second pointof attachment. Any current information flows shouldbe maintained during this process. The IETF proposedmobile IPv6 [4], this extension to IPv6 provides amechanism for a mobile device to leave its home networkand join a foreign network. When a mobile node entersthe foreign network it is issued with a care of address(COA) by a router in the foreign network. Once themobile agent has a COA, the mobile agent registersthis address with its home agent router. Data from thecorrespondent is routed to the home agent router, thentunnelled through the foreign agent router to the mobileagent using the COA. To reduce the increased latencyof tunnelling the data flow from the home agent tothe foreign agent. The home agent router can give thecorrespondent node the new COA of the mobile node.The tunnelling process could then be eliminated and thedata flow can go directly via the foreign router. Thisavoids the problems incurred through triangular routing[4].

The range of device mobility is usually referred to asmicro or macro mobility. If a node moves within accesspoints of the same domain or area of administrativecontrol this is referred to as micro mobility. Alternativelyif a node leaves its administrative area to join a separatedomain this is referred to as macro-mobility. Mobility

can be managed in two ways, network or device centric.Network centric mobility provides transparent mobilityto the mobile device. Whereas in mobile centric schemesthe mobile node initiate and have some control overthe establishment of tunnels and transmission of controlmessages [5].

A number of protocols have been suggested toreduce signalling latency for route establishment. FastHandovers for Mobile IPv6 (FMIP) [6] uses layer2 triggers to initiate the process of obtaining a newcare of address. Past Association-FMIP (PA-FMIP) [7]uses past association patterns to replace/augment thetriggers provided by layer 2. This results in a reductionof packet loss by 45% compared to Mobile IP. Withthe increasing use of Multi Protocol Label Switching(MPLS) to provide Quality of Service (QoS) in IPnetworks, Mobile MPLS [8] has been developed toreduce signalling overhead in micro mobility scenarioswhere MPLS is in use. Mobile MPLS proposes theuse of a Label edge gateway router, to create labelswitched paths that can take advantages of the highspeed of link establishment and QOS available to MPLStechnology. HAWAII [9] is a domain based approachessentially providing micro mobility within the ASdomain of the edge network. HAWAII updates selectedrouters as to the current connectivity of the mobiledevice by inserting host based entries for the mobilenode. Packets addressed to the IP of the mobile nodewould not follow the prefix for the subnet, instead thepacket would be routed by the individual route for thesingle device. Obviously this scheme would not scaleany higher than the Internet fringe and has a fall backmechanism using Mobile IP to cope with macro mobility.

There are a number of issues with curent node mobi-lity protocols, including:

• Increased signalling overhead when instigating theIP-IP tunnel

• Tunnel overhead especially for power constraineddevices with low battery and processor capabilities.

• Frame restrictions that are present in 802.15.4 whenusing 6LOWPAN would make encapsulation diffi-cult.

• If correspondents use COA it may result in subop-timal route as tunnel may be faster.

• Using new COA could cause firewalls to blocktraffic from the correspondent.

• Numerous single points of failure between mobileand home agent.

• Mobile centric mobility management would putadditional burden on constrained low power devices.

• Micro Mobility schemes rely on L2 triggers or

MPLS. To conserve battery resources on the mobiledevice, it may be better to use network centricmanagement for micro mobility.

There are a number of issues that need to be addressed inrelation to node mobility. Many of the issues reappearwhen we look at network mobility. One of the majorchallenges is reducing the overhead of tunnelling IP inIP, especially when we look to implementation on lowpower devices.

III. NETWORK MOBILITY

Mobile nodes that are part of a wireless domain willat some point, leave the wireless coverage area of thatdomain e.g. A Personal Area Network (PAN) leaving ahome area network. If there is a second domain that themobile node can associate with, packets can be routedthrough the new connection. These mobile devices couldbe treated as a number of individual nodes (node mobi-lity), however not all nodes will have the same power,radio or processing capabilities. These factors will limitthe extent of their wireless coverage range and theirability to maintain Mobile IP state. Setting up tunnelsfor each mobile device as required by mobile IP, wouldalso have an adverse impact on both home and foreignagent. To alleviate these issues, all nodes that are withindirect range of each other, or in some cases via Layer2 mesh routing, should be treated as a collective entity.This entity should then have the capability of attachingto other networks to obtain connectivity, that is betterthan its current level of connectivity.

The formation of mobile network segments can eitherbe achieved as an extension of the existing edge Inter-net connectivity, as can be seen in the IETF NetworkMobility (NEMO) standard. Alternatively we can applythe concepts of MESH ad hoc Networking and coreInternet principles as proposed by Chau et al in [10]. Theremainder of this section will review both approaches.

A. NEMO

The IETF have developed Network Mobility basic pro-tocol support (NEMO) [3]. This protocol is an extensionto Mobile IP and operates in a similar way. The mobilenetwork has a fixed router and as this network movesto a wireless zone with a different network prefix, therouter in the mobile network will acquire a COA fromthe foreign router. The Mobile router will then contactthe home router with the COA and the subnet of mobilenodes that it has connectivity with. The Home Routerwill perform packet-in-packet encapsulation to forwarddata from external correspondents to the mobile router

Figure 1. Nested NEMO

(Figure 1). The packets are then be forwarded to thenodes that are part of the mobile network. The mobilenodes have no knowledge of their mobility, responsibilityfor managing mobility rests with the mobile router. If anumber of mobile nodes are separated from their mobilerouter and home network, they would need to establishindividual mobile IP connections back to their homenetwork. As such there lacks a mechanism for the nodesto elect a new mobile router and form a new network ifa mobile network splits.

There are a number of issues with the IETF NEMOstandard, including:

• Encapsulating data from the Home Router to theMobile router will result in latency.

• The mobile router will incur significant radio ac-tivity resulting in battery use and consumption ofscarce on board resources.

• If used in a sensor network the overhead of tun-nelling could consume the memory and processingresources of the constrained device.

• There is no detection of domain fragmentation orsubsequent convergence.

• Lack of fair election of mobile gateways to supportsubsequent communication with home agent.

• Mobile networks would need to establish transitivetrusts with the networks that they form relationshipswith. Where multiple possible networks exist QoSwould need to be used to identify a preferredpartnership arrangement.

• Protecting the mobile network segment from attack

would be complex with no predefined roles orpoints of defence and a changing network topology.Traditional perimeter firewalls would need to bemore adaptable.

• If an access point receives multiple router solicita-tions. There is no mechanism to predict associationdue to past access patterns. This would enable aheadof time association and to prepare access point loadbalancing.

It is expected that networks will not only be mobile onthe Internet’s edge, but the greatest mobility will occurwithin the fringe of the Internet. This pattern of mobilitywill require NEMO networks to nest within each otherto provide access back to the edge Internet. NestingNEMO networks creates a pinball routing effect [11].This occurs when a 2nd mobile router obtains a careof address from the 1st mobile router. The 2nd mobilerouter must communicate through the 1st Mobile routershome agent. This creates a tunnel in tunnel overhead andsub optimal routing.

To reduce the tunnel in tunnel overhead, a number ofadditions to the NEMO standard have been suggested.NEMO+ [12] uses three separate components treediscovery, network in node advertisement and reverserouting header. Tree discovery enables mobile routersto exchange information to create a tree topology fromthe Internets edge, sharing reachability informationfor all of the nested networks. Network in nodeadvertisements maintains routes to all networks withinthe nested NEMO. Reverse routing header providesoptimal routing for all nodes in the nested NEMO tothe Internet.

Issues relating to NEMO in Nested configurationinclude:

• Nested NEMO inherits all of the problems associa-ted with Mobile IP.

• Deep nested fragmented networks can generate suboptimal paths due to the required tunnelling.

• Additions to NEMO can resolve sub optimal pathsat the expense of additional messaging and globalknowledge of all nested NEMO.

• Tree topology results in single point of failure, aswell as the overloading of mobile routers higher upthe tree, with both traffic and control messages.

• Nodes are unable to communicate between net-works where no connection to the edge Internet ispresent.

Deep nesting of NEMO networks will cause many issues,one main concern is as depth increases, there will begreater reliance on those nodes at the top of the tree.

A single failure close to the edge Internet could resultin a large number of disconnected nodes. Congestion atthis point in the topology would also have a detrimentaleffect on network performance.

B. Inter Mesh node mobility

In their paper titled "Inter- and Intra-Domain RoutingInteractions for MANETS" Ford et al [13] describe howin a battle field scenario multiple administrative domainswill be needed to cover the battlefield. These may beseparate coalitions each using their own internal protocolpossibly using legacy equipment as well as differentadministrative policy.Mobile mesh 6LOWPAN [14] networks operating on thefringe of the Internet, may need to route though a numberof other equally mobile networks to obtain connectivityto the edge Internet. Individual mesh autonomous net-work domains may split to form non contiguous areas.This differs from NEMO as it defines each mobilenetwork as an autonomy as apposed to a branch in acollective tree.In the core Internet, routing between separate adminis-trative domains is accomplished with Border GatewayProtocol (BGP) [15]. However BGP is designed for anon mobile network, where AS domains would remaincoherent. In a mobile network it would be likely to seeAS domains split into sub domains (figure 2). In thiseventuality BGP loop detection would disregard routesto one of the sections, as the AS number would appeartwice in routing advertisements.

Internal BGP uses the internal routing protocol of theAS to deliver packets to other designated fixed BGPgateway routers. In a mobile mesh environment, allnodes are routers and it may not by reasonable to assumethat a fixed subset of mesh devices could perform theBGP gateway function. This could be for reasons ofbattery use or device capability. Routing policy wouldalso need to be dynamic. On the Internet, autonomousnetworks establish peering relationships, often involvinghigh bandwidth data path arrangements, these relation-ships are not dynamic.

Inter Domain Routing for MANET (IDRM) [10] isa proposed first step solution to the mobile AS MESHproblem. Each IDRM AS has a number of set gatewaynodes that use an internal routing protocol to maintaininternal communication. Beacons are used to detect asplit in the MANET AS, if there is a split within theAS a new MANET ID is generated. Due to the arbitrarypartition IDRM relies on communicating a membershipdigest of the nodes within the AS instead of using theIP and prefix advertisement found on the Internet.

Issues relating to inter mesh mobility include:• Efficient election of domain gateway routers based

on the capability of nodes within the mobile net-work.

• Generation of MANET ID requires that all meshare in contact with one and other to minimise therisk of ID duplication. This would be unfeasible forlarge fragmented MESH networks.

• Advertising all nodes that exisit within an auto-nomous domain could lead to increasing networkadvertisment overhead. This scheme would fail inlarge internetnetworks with large intra node counts.

• Policy would need to be bound to AS id. If this ischanging how do we disseminate policy? Should afragmented AS carry the same policy as a wholeAS? The policy may need to be dynamic based onmobility and fragmentation.

• The MANET may be part of a larger network withfixed components e.g. a car in a garage that isattached to a house. Protocols should be aware ofthe mobility or static nature of network segments.

• Detecting a split in the autonomous domain sothat new network segment can be formed withoutrelying on periodic broadcasts.

• Determining policy and establishing trusts with ad-jacent autonomous domains.

• Maintaining routing state whilst many devices arein low power networks will be sleeping. Possiblyresolved using a collaborative routing approach.

There are many issues relating to the adoption of MESHAutonomous networks. Especially if we wish to adapttheir use to target low powered devices and the 6LOW-PAN stack. To maintain such a complex state would bechallenging on current low power platforms.

IV. OPEN RESEARCH CHALLENGES

There is a requirement to improve current routingsystems to account for the mobility of nodes, and moreimportantly collections of nodes within the fringe Inter-net. To detail the problems with current approaches wewill consider the following scenario.

A person has a personal area network (PAN) thatconsists of a mobile phone an mp3 player a heart monitorand accelerometer. The person also has a desktop PCand an Internet tablet, these devices combine with thePAN to become a single autonomous area for the person(perAN). When the person is at home all of thesedevices are in the same radio range and the networkis unfragmented. The persons home has a number ofnetworked appliances that form a separate autonomous

Figure 2. AS MESH Topology change

network including a fridge, lighting control, heatingsystem and burglar alarm. One of the most importantcomponents of the home autonomous network (hAN), isthe Internet router that is linked to the ISP autonomousnetwork. The router in the hAN is referred to as theInternet edge. Both the hAN and perAN utilise a rangeof network technology including 802.15.4, 802.11 andBluetooth to interconnect devices.

The perAN while in the house would likely usethe hAN to provide Internet connectivity. Internally theperAN is likely to elect the desktop PC to performforwarding functions within the perAN to the hAS.The hAS would need to provide a custome-consumerrelationship to the perAN for Internet access, in the sameway the ISP provides a customer-consumer relationshipto the hAS. The perAN could choose to set up a mutualpeering relationship with a neighbours hAS, to providean alternative route if the current hAS looses Internetconnectivity. This relationship would also provide theneighbour hAS, with the possibility of routing through inthe event of any internal failure within its own AS. Thisaspect of cooperation is lacking in current standards.

When the person leaves the house with the the PANdevices, the mobile phone could become this perASfragments mobile router through an election process. Ifthe person gets into a taxi, that is itself an autonomousnetwork that has a peering relationship with a vehicularad hoc AS network. What is the best point of attachment,the mobile phone or the taxi vehicular network? Youmight say the taxi but there are other factors to consider.Should the private information from the heart monitorbe transmitted over the vehicular network. If the personthen exits the taxi and forgets his phone the remaining

PAN components would need to elect a new router andcreate a new fragment perAS until the connectivity tothe phone or desktop PC is restored. The perAS currentlyconsists of three perAS fragments that should be capableof maintaining communication despite the fragmentationprocess. Although this communication may need to berestricted due to nested policy and firewall restrictionsin place between the fragments.

Existing standards fail to accommodate for thisfluidity or fragmentation of node relationships. Existingrouting standards have been developed to account forasynchronous access to the Internet in the case ofNEMO and have a number of issues relating to theapplication within the 6LOWPAN stack on low powerdevices.

To support the amalgamation of these technologiesand to support the application of the Internet of things anumber of research problems are apparant.

1) The interconnection of autonomous sections willneed to improve with the emphases on the seamlessrouting through deep nested meshes minimisingsingle points of failure.

2) Improved methods for establishing the formationof an autonomous network and the detection of anynetwork fragmentation are required.

3) Routing protocols should not work in isolationinside autonomous zones, instead reacting to offersof assistance in creating path improvements.

4) Security of the autonomous system routing domainshould be controlled through the partitioning ofrouting tables and the formation of dynamic bor-ders.

5) Reliance on single head nodes is not sufficientand autonomous mobile zones should be able toelect gateways based on a number of metrics andsurvive the splitting of AS into separate sectionswhilst maintaining whole AS connectivity. It maybe reasonable to suggest that multiple gateways arerequired as is found in BGP.

6) Traffic patterns in the Internet of things require amulti topology scheme as shown in the ROLL pro-tocol draft [16]. Extensions to traditional routingalgorithms do not meet the differing traffic flowsfound in sensor and other mobile applications. Anyrouting protocol suggested for Internet of thingsmust take into account these types of devices.

7) From the smallest of sensors to the most po-werful routers, distribution of complexity whencalculating routes should be considered. Devicesthat awaken should be provided stable commonlyrequested routes.

8) The introduction of prefixing should be investiga-ted as to reintroduce the ability for routing tables tosummarise routes as is found in the core Internet.

9) Routing protocols need to a adapt to the differinglevels of mobility through individual and collec-tive mobility patterns, building up knowledge andreacting more intelligently to their current state.

10) Traditional methods of detecting mobility such assignal strength will need to be augmented withother sensory input e.g. Accelerometer.

The proposed routing protocol should account for nodeand network fragmentation whilst maintaining connec-tivity. Nodes should form an autonomous system thatmaintains its own internal state and routing security.Autonomous systems should offer assistance to otherautonomous systems to establish dynamic peering re-lationships as can be found on the Internet core. Theresulting connectivity pattern would results in a meshoverlay mesh topology.

Any developments must account for the varied ca-pabilities of the devices found in the domain and iffound to be appropriate, operate a distributed approachto topology maintenance. For example if a mesh auto-nomous system advertises a path to another mesh auto-nomous system and that autonomous system is asleepthe route to that system should not be removed andany advertisements directed towards that network shouldbe processed cooperatively until the autonomous systemawakens. The autonomous system can then be providedcurrent network state without having to flood the networkfor updated routing information.

V. CONCLUSION

We have identified a number of issues relating tothe use of current mobility standards when looking tothe future requirements of mobile nodes in the fringeInternet. The Internet of things will drive requirementsbeyond the capability of current and proposed protocols.The next generation fringe mobile network will need tobe more dynamic and supportive of the autonomy thatwill exist, between the many devices that we will interactwithin the course of our daily routines. To providecomplete connectivity domains will need to fragmentedand coalesce, whilst providing peering relationships withother autonomous domains. Domains should recognisingthe limitations of the devices that form that autonomy,and fair election procedures should take place to informdevices roles. Intelligent peering will permit autonomoussections to interact and provide whole connectivity. Thefuture research challenges are extensive and complex,but realising them will help to fulfil the requirements ofubiquitous connectivity.

REFERENCES

[1] L. Gao, “On inferring autonomous system relationships inthe Internet,” IEEE/ACM Transactions on Networking, vol. 9,pp. 733–745, Dec. 2001.

[2] C. Perkins, “Mobile ip,” Communications Magazine, IEEE,vol. 35, pp. 84 –99, may 1997.

[3] V. Devarapalli, R. Wakikawa, A. Petrescu, and P. Thubert,“Network mobility (NEMO) basic support protocol.” InternetRFC 3963, Jan. 2005.

[4] J. A. D. Johnson, C. Perkins, “Mobility support in ipv6,” RFC3775, June 2004.

[5] E. Zagari, R. Prado, T. Badan, E. Cardozo, M. Magalhaes,J. Carrilho, A. Berenguel, D. Moraes, T. Dolphine, T. Johnson,and L. Westberg, “Design and implementation of a network-centric micro-mobility architecture,” in Wireless Communica-tions and Networking Conference, 2009. WCNC 2009. IEEE,pp. 1 –6, 5-8 2009.

[6] R. Koodli, “Fast handovers for mobile ipv6,” RFC 4068, July2005.

[7] A. Bergh and N. Ventura, “Pa-fmip: a mobility predictionassisted fast handover protocol,” in Military CommunicationsConference, 2006. MILCOM 2006. IEEE, pp. 1 –7, oct. 2006.

[8] R. Langar, S. Tohme, and G. Le Grand, “Micro mobile mpls:a new scheme for micro-mobility management in 3g all-ipnetworks,” in Computers and Communications, 2005. ISCC2005. Proceedings. 10th IEEE Symposium on, pp. 301 – 306,june 2005.

[9] R. Ramjee, K. Varadhan, L. Salgarelli, S. R. Thuel, S.-Y. Wang,and T. La Porta, “Hawaii: a domain-based approach for sup-porting mobility in wide-area wireless networks,” IEEE/ACMTrans. Netw., vol. 10, no. 3, pp. 396–410, 2002.

[10] C.-K. Chau, J. Crowcroft, K.-W. Lee, and S. H. Wong, “Inter-domain routing for mobile ad hoc networks,” in MobiArch ’08:Proceedings of the 3rd international workshop on Mobilityin the evolving internet architecture, (New York, NY, USA),pp. 61–66, ACM, 2008.

[11] E. Perera, V. Sivaraman, and A. Seneviratne, “Survey onnetwork mobility support,” Mobile Computing and Communi-cations Review, vol. 8, no. 2, pp. 7–19, 2004.

[12] B. McCarthy, M. Jakeman, C. Edwards, and P. Thubert, “Proto-cols to efficiently support nested nemo (nemo+),” in MobiArch’08: Proceedings of the 3rd international workshop on Mobilityin the evolving internet architecture, (New York, NY, USA),pp. 43–48, ACM, 2008.

[13] A. Ford, S. Wong, and C. Chau, “Inter-and Intra-Domain Rou-ting Interactions for MANETs,” in Proceedings of the SecondAnnual Conference of the International Technology Alliance,vol. 1, pp. 1–3, 2008.

[14] G. Mulligan, “The 6loWPAN architecture,” in EmNets (C. J.Sreenan, ed.), pp. 78–82, ACM, 2007.

[15] K. Butler, T. Farley, P. McDaniel, and J. Rexford, “A surveyof bgp security issues and solutions,” Proceedings of the IEEE,vol. 98, pp. 100 –122, jan. 2010.

[16] P. T. E. T. Winter (Ed.) and the ROLL Team, “Rpl: Ipv6 routingprotocol for low power and lossy networks.,” Internet Draftdraft-ietf-roll-rpl-07, work in progress, 2010.