COMP NW - Performance MIPv6 FMIPv6 HMIPv6 and Their Combination-preprint

A Performance Comparison of Mobile IPv6, HierarchicalMobile IPv6, Fast Handovers for Mobile IPv6 and their

Combination

Xavier Perez-Costaa Marc Torrent-Moreno ab Hannes Hartensteinab

[email protected]{Marc.Torrent-Moreno,Hannes.Hartenstein}@tm.uka.deaNetwork Laboratories, NEC Europe Ltd., Heidelberg, Germany

bUniversitat Karlsruhe (TH), Computer Center and Institute for Telematics, Germany (new affiliation)

Mobile IP, the current IETF proposal for IP mobility support, represents a key element forfuture All-IP wireless networks to provide service continuity while on the move within amulti-access environment. We conducted a performance evaluation of Mobile IPv6 and itsproposed enhancements, i.e., Fast Handovers for Mobile IPv6, Hierarchical Mobile IPv6and our proposed combination of them, using the network simulator ns-2 for the case ofa ‘hot spot’ deployment scenario. The simulation scenario comprises four access routersand up to 50 mobile nodes that communicate in accordance withthe IEEE 802.11 wirelessLAN standard. The study provides quantitative results of the performance improvementsobtained by the proposed enhancements as observed by a single mobile user with respect tohandoff latency, packet loss rate and achieved bandwidth per station. As a complementarypart of the study, the signaling load costs associated with the performance improvementsprovided by the enhancements has been analyzed. The simulation environment allowed usalso to investigate the behavior of the protocol in extreme cases, e.g., under channel sat-uration conditions and considering different traffic sources: CBR, VoIP, Video and TCPtransfers. While some simulation results corroborate the intention of the protocols specifi-cations, other results give insights not easily gained without performing simulations. Thisstudy provides a deep understanding of the overall performance of the various protocols andsupports the design process of a Mobile IPv6-based network when a decision of whetherit is appropriate to implement any of the proposed Mobile IPv6 enhancements has to bemade.

I. Introduction

The fast Internet evolution together with the enormousgrowth in the number of users of wireless technologieshas resulted in a strong convergence trend towards theusage of IP as the common network protocol for both,fixed and mobile networks. FutureAll-IP networkswill allow users to maintain service continuity whilemoving through different wireless systems.

The IETF working group in Mobile IP is proposingMobile IPv4 (MIPv4) [11] and Mobile IPv6 (MIPv6)[21] as the main protocols for supportingIP mobil-ity. Various enhancements to the MIPv6 base proto-col have been already proposed since it is believed thatin certain cases Mobile IP could result in a poor per-formance. For environments where the mobile nodescould change its point-of-attachment frequently andthe standard Mobile IP protocol could result in a highsignaling load as well as high handoff latency andpacket losses, micro-mobility protocols, as they arecommonly referred to, have been proposed [12]. Hi-

erarchical Mobile IPv6 [20] is the current IETF IPv6micro-mobility proposal. Additionally, for applica-tions that could suffer from long interruption timesdue to handoffs, Fast Handovers for Mobile IPv6 [19]has been designed.

This paper investigates the impact of various pa-rameters on the overall performance as experiencedby a single mobile node of a Mobile IPv6-based wire-less access network and compares the performanceobtained by the proposed enhancements, i.e., Hier-archical Mobile IPv6 (HMIPv6), Fast Handovers forMobile IPv6 (FMIPv6), or our proposed combinationof both (H+F MIPv6), with the performance of theMIPv6 base protocol.

We are primarily interested in quantifying thedegradation of quality of service a mobile user per-ceives during a handoff whenreceivinga data stream(e.g., video or voice over IP) and the signaling loadcosts associated with Mobile IPv6 and its enhance-ments. More specifically, we are interested in perfor-mance metrics like handoff latency, packet loss rate,

Mobile Computing and Communications Review, Volume 7, Number 4 5

obtained bandwidth per station and signaling load.Moreover, the impact of different traffic sources isstudied: CBR, video, VoIP and TCP transfers. Thescenario chosen for this study resembles a ‘buildingblock’ of a potential wireless LAN ‘hot spot’ deploy-ment, as one of the possible wireless access networksin anAll-IP network. It comprises four access routersand up to 50 mobile nodes that communicate in accor-dance with the IEEE 802.11 wireless LAN standard.We study the performance metrics as observed by onesingle mobile node that either moves deterministicallyor randomly while the other mobile nodes move ran-domly all the time providing realistic ‘interference’with respect to the observed mobile node. The mobil-ity model used for the random movement is the Ran-dom Waypoint Mobility Model [7]. We consider theimpact of different parameters like number of mobilenodes, handoff rate of the observed MN, number ofcorrespondent nodes, wired link delay, and specificprotocol options over the various performance met-rics. Due to the complexity and broadness of the re-quired study, simulation was chosen as the most suit-able analysis method. As simulation tool we used thenetwork simulatorns-2.

Previous work on simulative evaluations of MobileIP almost exclusively dealt with IPv4 networks. Be-cause of the significant differences between MobileIPv6 and Mobile IPv4, as outlined in the followingsection, results obtained for MIPv4 do not take overfor MIPv6. We have focused on IPv6 since we believeit will be the basis of the futureAll-IP networks as itcan be seen for example with the 3GPP decision ofadopting IPv6 as the only IP version for the IP-basedmultimedia subsystem (IMS).

Regarding Mobile IPv6, an analitycal study exclu-sively focusing on the HMIPv6 update signaling mes-sages frequency can be found in [8]. A protocoloverview of Mobile IPv6, HMIPv6 and FMIPv6 isprovided in [23] but the obtained results are restrictedto the case of the handoff latency for MIPv6, exclud-ing HMIPv6 and FMIPv6 and considering only the in-terference of up to 4staticusers. In [15] Mobile IPv6and its enhancements is studied but the results are lim-ited to TCP handoff latency and obtained bandwidthof a single user following a deterministic path with-out the interference of other users. Moreover, a keyaspect of IPv6, the Neighbor Discovery protocol, hasnot been implemented. We have implemented Neigh-bor Discovery since it has a relevant impact on theresults, as explained in Section II.A. In [15] a sim-ple aggregation of HMIPv6 and FMIPv6 is consideredwhereas in this paper we put forward a fullintegration

of both approaches, as described in Section II.D.In contrast to the related literature, in our previ-

ous [25, 26, 27, 22] and current work we performa detailed study of Mobile IPv6, Hierarchical MobileIPv6, Fast Handovers for Mobile IPv6 and their com-bination focusing not only on handoff latency but ona complete picture of the overall performance takinginto account a variety of performance metrics as wellas impacting factors. Moreover, while previous anal-ysis usually studied a single mobile node without theinterference of others, our work considers a more re-alistic scenario with up to 50 mobile nodes and ran-dom movement patterns. Our goal is not to deter-mine which protocol performs ‘best’ but to assess theperformance that can be expected for each protocol,broaden our knowledge of the reasons that influencethe difference in the performance and help in the de-sign decision of which is the best suited protocol for aspecific scenario.

The rest of the paper is organized as follows.Section II describes the basics of MIPv6, HMIPv6,FMIPv6 and our combined H+F MIPv6 approach. InSection III we describe the simulation setup. Perfor-mance aspects subject of interest are given in SectionIV. Simulation results are provided and discussed inSection V. Finally, Section VI summarizes the resultsand concludes the paper.

II. Mobile IPv6

Mobile IP supports mobility of IP hosts by allowingthem to make use of (at least) two IP addresses: ahome address that represents the fixed address of thenode and a care-of address (CoA) that changes withthe IP subnet the mobile node is currently attached to.Clearly, an entity is needed that maps a home addressto the corresponding currently valid CoA.

In Mobile IPv4 [11] these mappings are exclu-sively handled by ‘home agents’ (HA). A correspon-dent node (CN) that wants to send packets to a mobilenode (MN) will send the packets to the MN’s homeaddress. In the MN’s home network these packetswill be ‘intercepted’ by the home agent and tunneled,e.g. by IP-in-IP encapsulation [9], either directly tothe MN or to a foreign agent to which the MN has adirect link.

In MIPv6, home agents no longer exclusively dealwith the address mapping, but each CN can have itsown ‘binding cache’ where home address plus care-ofaddress pairs are stored. This enables ‘route optimiza-tion’ compared to the triangle routing via the HA inMIPv4: a CN is able to send packets directly to a MN

6 Mobile Computing and Communications Review, Volume 7, Number 4

when the CN has a recent entry for the MN in its cor-responding binding cache. When a CN sends a packetdirectly to a MN, it does not encapsulate the packetas the HA does when receiving a packet from the CNto be forwarded, but makes use of the IPv6 RoutingHeader Option. When the CN does not have a bind-ing cache entry for the MN, it sends the packet to theMN’s home address. The MN’s home agent will thenforward the packet. The MN, when receiving an en-capsulated packet, will inform the corresponding CNabout the current CoA.

In order to keep the home address to CoA mappingsup-to-date, a mobile node has to signal correspond-ing changes to its home agent and/or correspondentnodes when performing a handoff to another IP sub-net. Since in MIPv6 both, HA and CN, maintain bind-ing caches, a common message format called ‘bindingupdates’ is used to inform HA and CNs about changesin the point of attachment. Additionally, since the BUshave associated a certain lifetime, even if the MN doesnot change its location a BU to its HA and CNs isnecessary before the lifetime expires to keep alive theentry in the binding caches. In the rest of the paperthose BUs will be referred as periodic BUs. Bind-ing updates (BU) can be acknowledged by BU Acks(BAck).

In contrast to MIPv4, where signaling is done us-ing UDP, Mobile IPv6 signaling is done in exten-sion headers that can also be piggybacked on ‘reg-ular’ packets. To acquire a CoA in Mobile IPv6, amobile node can build on IPv6 stateless and state-ful auto-configuration methods. The stateless auto-configuration mechanism is not available in IPv4. Inour work, we assume stateless auto-configuration forall tests since with this mechanism it is not necessaryto contact any entity to obtain a new CoA, reducingthe handoff process duration. For more details on Mo-bile IPv6 see [21]. In the following, we briefly look atthe Neighbor Discovery [16] mechanism, one of themain differences when comparing IPv4 and IPv6.

II.A. Neighbor Discovery

Neighbor Discovery [16] is used by nodes to resolvelink-layer addresses and keep track of the reachabil-ity of their neighbors. Hosts use it as well to lo-cate routers on their link. The main difference is theIPv6 way of learning MAC addresses and the Neigh-bor Cache, previously ARP Cache, which can be setin five different states: Incomplete, Reachable, Stale,Delay and Probe.

A MN, when performing a handover, has to learnthe Access Router’s (AR) MAC address before be-

ing able to inform about the new point of attachmentvia the BUs. In IPv4 a MN runs the ARP processand has to wait until its completion, delaying thus theBUs transmission. On the other hand, the IPv6 Neigh-bor Discovery protocol optimizes this process obtain-ing the AR’s MAC address from the Router Adver-tisement. This results in the MN being able to sendthe BU without any delay after a handover and run-ning the neighbor unreachability detection process inparallel. However, in IPv4, after the ARP processis completed, MAC addresses on both sides are ob-tained. This is not the case for IPv6 where the ARthat has a packet to transmit to the MN must run theaddress resolution process to obtain the MN’s MACaddress. In fact, in the IPv6 case, when a MN learnsa node’s MAC address in a different way than theusual Request-Reply exchange or when it wants tosend a packet after some time without using the en-try, the neighbor unreachability detection has to belaunched to resolve the MAC address, but this is a oneway process (only one address is resolved). Note thatin both cases, addresses will be resolved in parallelwhile sending packets, no delay is added. Addition-ally, some channel utilization can be saved if confir-mation of reachability is received from upper layers.

II.B. Fast Handovers for Mobile IPv6

To reduce the service degradation that a mobile nodecould suffer due to a change in its point of attachmentFast Handovers for Mobile IPv6has been proposed[19]. During the IETF discussions regarding this pro-posal two different mechanisms have been described:anticipated and tunnel-based handover. Tunnel-basedhandover relies on link layer triggers to potentiallyobtain better results than Anticipated Handover, in-troducing though a link layer dependence that couldmake the solution unfeasible for some link layer tech-nologies. In principle, a link layer independent solu-tion would be a more desirable solution. Therefore,we have focused on the performance study of theAn-ticipated Handoverproposal, which is solely based onnetwork layer information.

Anticipated Handover proposes a ‘make-before-break’ approach. When a MN has information aboutthe next point of attachment to which the MN willmove, e.g., via reception of a Router Advertisementfrom a new AR (nAR), it sends a Router Solicita-tion for Proxy (RtSolPr) to the old AR (oAR) with anidentifier of the point of attachment to which it wantsto move. Once the oAR receives information that aMN wants to move to a nAR, it constructs a nCoAbased on the MN’s interface ID and the nAR’s subnet


prefix. It then sends a Proxy Router Advertisement(PrRtAdv) to the MN containing the proposed nCoAand the nAR’s IP address and link layer Address. Atthe same time, the oAR sends a Handover Initiate (HI)message to the nAR, indicating the MN’s oCoA andthe proposed nCoA.

Upon receipt of the HI message, the nAR first es-tablishes whether there is already an active NeighborCache entry for the proposed nCoA. If the nCoA isaccepted by the nAR, the nAR adds it to the Neigh-bor Cache for a short time period so it can defend it.The nAR then responds with a Handover Acknowl-edge (HAck), indicating that the proposed nCoA isvalid. Upon receipt of the HAck the oAR is preparedto forward packets for the MN to the nCoA. As soonas the MN received confirmation of a pending net-work layer handover through the PrRtAdv and has anCoA, it sends a Fast Binding Update (F-BU) to oAR,as the last message before the link layer handover isexecuted.

On receipt and validation of the F-BU, the oARresponds with a Fast Binding Acknowledgment (F-BAck), destined to the nCoA. The oAR waits for a F-BU from the MN before actually forwarding packets.On receipt of the F-BU, the oAR forms a temporarytunnel for the lifetime specified in the F-BAck, andthe F-BAck is sent through the tunnel to the MN onthe new link. When the MN arrives to the nAR and itslink layer connection is ready for network layer traf-fic, it sends a Fast Neighbor Advertisement (F-NA)to initiate the flow of packets that may be waiting forit. The nAR will deliver packets to the MN as soonas it receives an indication that the MN is already at-tached to it, usually receiving a F-NA from the mo-bile node. The oAR is responsible for forwarding anypackets that arrive for the MN under its oCoA afterthe MN has moved. Once the fast handoff process iscompleted, the MN will follow the MIPv6 normal pro-cedure of informing the HA and correspondent nodesabout its new location. For more details about FastHandovers for Mobile IPv6 see [19].

II.C. Hierarchical Mobile IPv6

It is a well-known observation that MNs movingquickly as well as far away from their respective homedomain or correspondent nodes produce significantBU signaling traffic and will suffer from handoff la-tency and packet losses when no extension to the base-line Mobile IP protocol is used. Hierarchical MobileIPv6 (HMIPv6) is a localized mobility managementproposal that aims to reduce the signaling load due touser mobility. The mobility management inside the

local domain is handled by a Mobility Anchor Point(MAP). Mobility between separate MAP domains ishandled by MIPv6.

The MAP basically acts as a local Home Agent.When a mobile node enters into a new MAP domainit registers with it obtaining a regional care-of address(RCoA). The RCoA is the address that the mobilenode will use to inform its Home Agent and corre-spondent nodes about its current location. Then, thepackets will be sent to and intercepted by the MAP,acting as a proxy, and routed inside the domain to theon-link care-of address (LCoA). When a mobile nodethen performs a handoff between two access pointswithin the same MAP domain only the MAP has to beinformed. Note, however that this does not imply anychange to the periodic BUs a MN has to sent to HA,CNs and now additionally to the MAP.

HMIPv6 presents the following advantages: it in-cludes a mechanism to reduce the signaling load incase of handoffs within the same domain and may im-prove handoff performance reducing handoff latencyand packet losses since intra-domain handoffs are per-formed locally. However, since the periodic BUs arenot reduced but the ones due to handoffs, the gain de-pends on the mobility of the mobile nodes. For moredetails on HMIPv6 the reader is referred to [20].

II.D. Hierarchical Mobile IPv6 plus FastHandovers for Mobile IPv6

In this section we describe our proposed combina-tion of FMIPv6 and HMIPv6 which was designed toadd up the advantages of both and provide additionalimprovements. In [20] a sketch on how to combineFMIPv6 and HMIPv6 is provided. However, someissues are left open as for example when should themobile node decide to perform the handoff. The mainideas of our approach and its differences with respectto a simple aggregation of the proposals described inthe above sections are as follows.

Our approach is based on two main observationsthat show that a simple aggregation of HMIPv6 andFMIPv6 would be inefficient. First, consider a MAPplaced in an aggregation router above the ARs in-volved in a handover. The usual fast handover pro-cess of forwarding packets from the oAR to the nARwould be inefficient in terms of handover latency sincepackets would traverse the MAP-oAR link twice andcould arrive disordered. On the other hand, if the en-tity responsible of establishing the redirection prior tothe handoff would be the MAP, then this inefficiencywould be removed. Therefore, in our approach, assuggested in [20], the entity performing the function-


ality of the Fast Handover process is the MAP insteadof the mobility agent in the old access router.

Second, note that with FMIPv6 the traffic is redi-rected when the oAR receives the F-BU but in ourcase if the mobile node would perform the handoffright after sending the F-BU to the MAP, all the pack-ets forwarded to the oCoA, during the period that theF-BU requires to arrive to the MAP, would be lost.Additionally, if the MN would perform the handoffright after sending the F-BU, it would not immedi-ately receive any redirected packet for the same rea-son, increasing the handoff latency and packet losses.As a solution, we propose to wait as long as possible(until connectivity is lost) for the F-BAck at the oldlink to start the handover. In this case we assure thatwhen we receive the F-BAck there are no packets lostsent to our oCoA and the ones redirected to our nCoAare buffered, i.e., no packet losses. Additionally, as-suming that the packets experience a similar delay inthe path between the MAP and the ARs involved inthe handoff, the reception of the F-BAck would act asa kind of synchronization packet telling us that newpackets are already waiting or about to arrive to thenew AR and therefore, the handover latency due tothe wired part would be almost removed.

Our approach requires, as in the case of FMIPv6,that the MN has some time from the moment it re-alizes that a handover should be performed until it isnecessary to perform it because of losing connectiv-ity to the current AR. In the cases where this is notpossible we apply the same recovery mechanisms asFMIPv6.

Addendum: During the preparation of this papera new internet-draft [14] appeared proposing a com-bination of HMIPv6 and FMIPv6 basically explain-ing in detail what was indicated in [20] but withoutthe proposed optimization of waiting at the old accessrouter for the F-BAck.

III. Simulation setup

The studied scenario was designed in order to be largeenough to provide realistic results but to be smallenough to be handled efficiently withinns-2. The cho-sen scenario, depicted in Figure 1, is composed bythe Home Agent and the Correspondent Nodes thatare connected via the ‘Internet’ (modeled by adjust-ing the link delayld) to a central router (CR). Four ac-cess routers (AR) –each one representing a differentIP subnet– are connected via two intermediate routers(IR) to the central router. When Hierarchical MIPv6is considered, the functionality of the Mobility An-

CN 1 HA

IR

AR1 AR2

IR

AR3 AR4

ld ld

2ms

2ms

CN N

CR

MN

IEEE 802.11

(MAP)

Figure 1: Simulation scenario

AR1

AR4 AR3

AR2

MN

MN

MN

MN

MN

MN

MN

MN

MN

MN

Figure 2: Access routers location

chor Point is placed on the central router and the CR,IRs, and ARs form the micro-mobility domain. At thestart of the simulation the mobile nodes are uniformlydistributed over the system area.

The access routers have been positioned in a wayto provide total coverage to an area of approximately700 × 700 square meters considering a transmissionrange of 250 meters, see Figure 2. The mobile nodesmove randomly within the coverage area followingthe random waypoint mobility model (RWP) [7]. Thismodel has been previously used mainly for ad-hocsimulations but it is well suited as well also for ourpurposes, more details are given in Section V. Aswireless medium the 2Mbps Wireless LAN 802.11DCF [5] provided byns-2[2] is used.

Within the micro-mobility domain each wired con-nection is modeled as a 5Mbps duplex link with 2msdelay. The ‘Internet’ connecting the central router andthe HA or CNs is modeled also as a 5Mbps duplex linkwith a default link delay (ld) of 10ms. In the simula-


tions, theld value has been varied to model various‘distances’ between the MNs and the HA and CNs.

While moving within the overlapping area, the mo-bile nodes are able to send/receive dataonly via theaccess router that corresponds to their current careof address. Technologies like 802.11 allow the mo-bile nodes gathering information about the neighbor-ing access routers, but do not allow to receive IP flowsat different frequency bands simultaneously from twoaccess routers, except for particular cases like havingan additional wireless interface.

In order to simulate a realistic case where a MN willreceive packets from the shared AR queue and wherea MN will also compete with other MNs and with anAR to access the channel, half of the MNs receive datafrom the CNs and the other half send data to the CNs.The CNs sending to the MNs introduce delay in theAR queue and the MNs sending to the CNs introducedelay in the wireless link. The study though focuseson the MNs receiving data from the CNs which are themost affected ones by the handoffs since the purposeis to analyze the degradation of the user experiencedquality of service due to mobility.

It is important to note the following fact that resultsfrom using a shared access: whenever we work closeto the maximum throughput of the channel, the MNsthat will first experience a reduction in their through-put will be the onesreceivingfrom the CNs. The rea-son is that these stations receive all the packets fromthe same station, i.e., the AR, sharing the access queueto the wireless channel, while the other MNssendingto the CNs do not share their access queue.

In our simulations we study the performance met-rics as observed by one single mobile node but af-fected by other moving mobile nodes. In most of thesimulations the observed mobile node follows a deter-ministic path while all other mobile nodes move ran-domly. This case allows for full control of the mobil-ity – and handoff rate – of the observed node whilethe interference of other nodes is still realistic due totheir random movements. As a second case we allowthe observed mobile node to move randomly, too. Bydoing this, mobility is less ‘controllable’ but randommovement effects – like going back and forth betweentwo ARs – can be analyzed. Thus, with both deter-ministic and random movements of the observed nodestudied separately, impact of the different parametersover the various protocols can be studied in a clear aswell as realistic way.

The first type of sources used in our simulationswill be UDP CBR sources. These sources provideconstant traffic where no acknowledgments are re-

quired. This kind of traffic is usually generated byreal-time applications and due to its deterministiccharacteristics, without recovery mechanisms, easesthe protocols study and comparison. Unless otherwisenoted, UDP CBR sources are used.

One of the applications expected to be used withMIPv6 is VoIP. We have implemented a VoIP modelbased on the one provided in [13]. The modelassumes silence suppression and models each voicesource as an on-off Markov process. The alternat-ing active on and silenceoff periods are exponen-tially distributed with average durations of 1.004s and1.587s. As recommended by the ITU-T specificationfor conversational speech [3], an average talk spurt of38.57% and an average silence period of 61.47% isconsidered. A rate of 88 kbps1 in on periods and 0kbps inoff periods is assumed for a voice source thatgenerates CBR traffic.

As streaming application for real-time video trafficwe have used a real H.263 [4] video encoding pro-vided by [18] (film: ”Star Trek: First Contact”) for atarget bit rate of 64 kbps. The obtained frame sizes(in bytes) of the individual encoded video frames areused as input for thens-2real-time video traffic appli-cation. Since these traces include only the raw pack-etized video, additional streaming protocol overheadhas been added. As in the case of VoIP sources weconsider a 12 byte RTP header plus 8 byte UDP headerand plus 40 byte IPv6 header as the streaming proto-col overhead.

TCP is the most widely used transport protocol. Wesimulate endless FTP sources to understand the im-pact of IP mobility on the congestion control mecha-nism of TCP.

The simulation code used for the experiments wasdesigned on top of INRIA/Motorola MIPv6 [1] codefor ns-2 [2] implementation. We have extended thecode with four main modules: Neighbor Discovery,Hierarchical Mobile IPv6, Fast Handovers for MobileIPv6 and their combination. The whole functionalitydescribed in Section II has been implemented.

IV. Performance metrics

The purpose of the performance comparison is toquantitatively evaluate the improvements that mobileusers would experience in a system using the proposedenhancements in comparison to the baseline MIPv6.

1Assume 8KHz 8 bits/sample PCM codec was used with20ms frame per packet. With 12 byte RTP header, 8 byte UDPheader and 40 byte IPv6 header, the size of each voice packetis 220 bytes. The bandwidth required will be (220 × 8)/20 ×

10−3=88kbps


The parameters to be studied are as follows:Handoff Latency:Handoff latency is defined for a

receiving MN as the time that elapses between the lastpacket received via the old route and the arrival of thefirst packet along the new route after a handoff. La-tency is an important parameter for delay sensitive ap-plications like video or VoIP. This packet drop periodwould result in a flickering image for a video appli-cation or in a noticeable disruption in the voice trans-mission for VoIP. We study handoff latency for CBRand video sources, for various values of link delaysld

and for an increasing number of mobile nodes.Packet Loss:Packet loss is defined for a receiving

MN as the number of packets lost during the hand-off. While one usually assumes that packet losses aredirectly proportional to latency it will be shown thatthis is not true in some cases. We have studied sepa-rately the packet losses due to the address resolutionprocess, the packet losses in the old access router andthe packet losses in the Home Agent. We study packetlosses for CBR and VoIP sources, for various valuesof link delaysld, for an increasing number of mobilenodes as well as for random movements.

Signaling Load: The signaling load is definedfor MIPv6 and HMIPv6 as the number of BUs andBAcks received during the simulation. Addition-ally, in the FMIPv6 and H+F MIPv6 case the BUs,BAcks, PrRtAdv, PrRtSol, F-NA, F-BU, F-BAck, HIand HAck signaling messages are also considered.We study the signaling load for various handoff rates(number of handoffs per minute) and different num-ber of correspondent nodes, and differentiate betweensignaling load within and outside the micro-mobilitydomain.

Bandwidth per Station:We study the probability toobtain the required bandwidth and the correspondingexpected variance for CBR and TCP sources for anincreasing number of competing stations.

Note that the whole set of performance metrics havebeen obtained for each scenario but only the most rel-evant results have been included.

V. Performance evaluation & discus-sion

With our ns-2 simulations we study the parametersexplained in Section IV for the scenario described inSection III. Unless stated otherwise, we analyze thedegradation of the performance metrics from the pointof view of a single mobile node that follows ade-terministic path while all other mobile nodes in thesystem follow the random waypoint mobility (RWP)

0

0.5

1

1.5

2

2.5

3

3.5

4

10 15 20 25 30 35 40 45 50

Late

ncy

(s)

Number of Stations

FMIPv6MIPv6

H+F MIPv6HMIPv6

0

0.02

0.04

0.06

0.08

0.1

10 15 20 25 30

Figure 3: Impact of number of stations on handoff la-tency

model. The RWP model is well-suited to representmovements of mobile users in campus or ‘hot spot’scenario at moderate complexity. When no othervalue is indicated, all the simulations have been per-formed with a maximum speed of 5m/s.

To obtain accurate results we have chosen a UDPprobing traffic from the CN to our specific mobilenode of 250 bytes transmitted at intervals of 10 ms.The other mobile nodes create background trafficsending or receiving data at a rate of 32 kbps.

All simulations have a duration of 125 seconds witha 5 seconds warm-up phase. Each point in the fol-lowing graphs represent the average of at least 100simulations. The sample size necessary to achieve aconfidence interval of 99% with respect to the aver-age value has been selected as indicated in [17]. Thisrequired in some cases to perform up to 1000 simu-lation runs, e.g., in the 50 mobile nodes or randommovement case.

We assume a system where mobile nodes use theIPv6 stateless address auto-configuration feature [24]performing Duplicate Address Detection (DAD) inparallel to avoid the introduction of an additional de-lay to the handoff process. Note that the delay intro-duced by DAD would be too time consuming resultingin a noticeable disruption of the service.

V.A. Impact of number of stations

We present here the results of the impact of the num-ber of competing stations on the following parame-ters: handoff latency, packet loss, obtained bandwidthand the fast handoff process probability of success.

The studied MN performs 4 handoffs during a sim-ulation run moving at 10 m/s from center to centerof the ARs’ coverage areas until it reaches again thestarting point. The values represented in the graphs


0

1000

2000

3000

4000

5000

10 15 20 25 30 35 40 45 50

Pac

ket L

osse

s

Number of Stations

FMIPv6HMIPv6

H+F MIPv6MIPv6

0

10

20

30

40

50

60

70

10 15 20 25 30

Figure 4: Impact of number of stations on packetlosses

0

500

1000

1500

2000

2500

10 15 20 25 30 35 40 45 50

Pac

ket L

osse

s N

D

Number of Stations

MIPv6FMIPv6HMIPv6

H+F MIPv6

0

5

10

15

20

25

10 15 20 25 30

Figure 5: Impact of number of stations on packetlosses in the Neighbor Discovery resolution queue

correspond to the analyzed MN.Figures 3 and 4 show the increase in handoff la-

tency and packet losses due to an increase in the num-ber of MNs sharing the wireless channel. We can ob-serve that up to 20 MNs the results are as expectedconsidering that for a small number of mobile nodes,e.g. 20 or below, the dominating factor of the handofflatency is the wired delay. HMIPv6 latency outper-forms standard MIPv6 one since the wired ‘distance’in order to update the entity that forwards packets tothe mobile node is always shorter. FMIPv6 outper-forms standard HMIPv6, since the MN prepares thehandoff in advance and thus, after a handoff, does nothave to wait for the oAR to be updated to start re-ceiving packets again. With FMIPv6 packets are redi-rected by the oAR to the nAR through the wired linkand therefore only this delay is noticed. H+F MIPv6performs better than all the other solutions since, asexplained in Section II.D, when the MN receives theF-BAck from the MAP indicating that the handoffshould be performed, the re-directed packets are al-ready waiting in the new AR.

0

500

1000

1500

2000

2500

3000

3500

4000

10 15 20 25 30 35 40 45 50

Pac

ket L

osse

s H

A

Number of Stations

H+F MIPv6HMIPv6FMIPv6

MIPv6

0

1

2

3

4

5

10 15 20 25 30

Figure 6: Impact of number of stations on packetlosses at the HA

0

0.2

0.4

0.6

0.8

1

10

Number of Stations

Nu

mb

er

of

occ

urr

en

ces

(%)

20 30

H+F MIPv6 Partial Completion

H+F MIPv6 Full Completion

FMIPv6 Partial Completion

FMIPv6 Full Completion

40 50

Figure 7: Fast Handover process success histogram

An exceptional case can be observed for 30 MNswhere MIPv6 shows a slight better performance thanHMIPv6. Due to the encapsulation that HMIPv6 al-ways does from the MAP to the current point of at-tachment we have a higher load on the channel, i.e., 40additional bytes per packet, and thus HMIPv6 reachesearlier saturation conditions, increasing the wirelessdelay that now dominates over the wired one. This dif-ference can not be noticed in the H+F MIPv6 case be-cause although we have the same encapsulation prob-lem, the higher load in the channel does not have a di-rect impact on the handoff performance due to the fasthandover mechanism that prepares the handover in ad-vance and re-tries up to three times. However, whenthe wireless delay becomes very high due to saturationin the channel, e.g., 40-50 stations case, we have againa better performance of HMIPv6 in comparison withMIPv6 due to two reasons. First, in the HMIPv6 casethe BU to the MAP is sent right after attaching to thenew link while MIPv6 sends a BU to the HA beforethe one to the CN, i.e., introducing an additional wire-less delay. This difference could be removed send-ing the BU first to the CN and then to the HA. Sec-ond, while the BAcks to HA and MAP are mandatory,


100

120

140

160

180

200

220

10 15 20 25 30 35 40 45 50

Ban

dwid

th (

kbps

)

Number of Stations

HMIPv6H+F MIPv6

MIPv6FMIPv6

Figure 8: Impact of number of stations on bandwidthobtained by observed MN

the BAck to the CN is optional. In our implementa-tion BAcks to CN BUs are not sent to avoid additionaloverhead and because in case of the BU being lost, theMN will re-send it again when receiving a data packetfrom the HA instead of directly from the CN. Underhigh saturation channel conditions the probability of aBU to be lost is higher, therefore, when using standardMIPv6 if a BU to the CN is lost2, it is not retransmit-ted, increasing significantly the latency value. On theother hand, when the BAck from the MAP is not re-ceived, the BU will be retransmitted.

Although the Fast Handover protocol is designedto minimize packet losses and latency during a hand-off, we can observe a worse performance with re-spect to MIPv6 when saturation arises. To under-stand this behavior a few factors must be considered.In the scenarios with 40 or more MNs the load inthe wireless channel is high, resulting in a channelwith a long access time and high collision rate. Ifwe take a look at the packets lost at the neighbor dis-covery resolution queue3 (ND), Fig.5, we can see thatthey are higher when FMIPv6 is not used (they dou-ble with 50 MN). Those packets, that are dropped inthe ND entry queue, are not sent through the wirelesschannel, which results in a lower channel saturationand, what is more important, a shorter access delay.In the FMIPv6 scenario though, the nAR learns thelink layer address of the MN before having to send apacket to it (via the reception of the PrRtSol by theoAR which triggers the HI-HAck handshake) even if

2Note that IEEE802.11 realizes when a packet was not cor-rectly transmitted over the wireless medium due to the lack of aMAC layer acknowledgment and re-tries the transmission a cer-tain number of times before discarding it (8 in our case)

3During the address resolution process only a small amount ofpackets are buffered for the same destination address, e.g., threein our implementation [16]

the FMIPv6 process has not been successfully per-formed. Therefore, the AR will send packets throughthe wireless medium without waiting for the addressto be confirmed, once the F-NA has been received, in-troducing a higher load on the channel.

H+F MIPv6 and HMIPv6 present, under saturationconditions, similar packet losses results since the pro-cess to update the MAP and afterwards HA and CNabout the new point-of-attachment is the same for bothapproaches (see Figure 6). They show the worst per-formance in packets lost at the HA, which is actuallya good measure of whether the route updating mecha-nisms are working properly. Packet are lost at the HAonly when the BU lifetime of both, CN and HA, hasexpired. As it can be seen in the figure, the higherload for 30 or more MNs produces a higher rate ofpacket losses at the HA. Which actually are most ofthe packet losses experienced in the H+F MIPv6 andHMIPv6 case. The reason is that the MN has to waitfor the MAP’s BAck to send the BUs to HA and CN,what can take a long time when the wireless channelis highly congested, resulting in the expiration of theBU lifetime (10s in our experiments) of the HA andCN. H+F MIPv6 obtains a slight higher HA packetloss rate due to its additional signaling load (see Sec-tion V.B). These higher packet losses in the HA arecompensated by lower packet losses due to NeighborDiscovery. Note that if the first signaling message ofthe fast handover procedure (PrRtSol) arrives at itsdestination, triggering the HI, the nAR will alreadyhave the link layer address before having to forwarddata packets to the MN, which explains the slight dif-ference between H+F MIPv6 and HMIPv6. Anotherremarkable aspect of the ND packet losses graph isthe big difference in saturation conditions between theprotocols that use a hierarchical approach and the oth-ers. The ND procedure is triggered by the first packetreceived in the nAR, the BAck from the MAP. Using ahierarchical approach and under saturation conditionsthe BAck is not always immediately followed by datapackets (because the HA and CN have not been up-dated on time and packets are being dropped in theHA) providing some additional time to the nAR to re-solve the link layer address.

Although all the differences (either in ND or HAPacket Losses) described for the congestion case, wecan observe that once the saturation level has beenreached by all the protocols, if we increase the num-ber of MNs the packet losses tend to converge, sincefor all cases the wireless channel presents a high colli-sion rate and long channel access time reducing thus,the impact of the differences between the approaches.


Figure 7 perfectly shows the saturation of the channeldepending on the number of MNs. Up to 30 MNs thewireless channel conditions allows for a proper com-pletion of the fast handoff process. However, for ahigher number of MNs the probability of the processsuccess decreases dramatically. We have differenti-ated between two cases: full completion of the fasthandoff process and partial completion, i.e., the redi-rection of the traffic from the oAR to the nAR has beenestablished. We believe that the latter case is a signif-icant value since it means that the F-BAck packet hasbeen lost but not the previous FMIPv6 correspondingmessages, resulting in a smoother handoff comparedto MIPv6. H+F MIPv6 presents a better performancethan FMIPv6 since most of the packet are lost in theHA reducing the load introduced in the wireless chan-nel compared to FMIPv6.

Figure 8 corresponds to the bandwidth obtained byour specific mobile node. As we can see, the band-width correlates almost perfectly the results shownfor packet losses. The slight difference between bothgraphics (in the 40 and 50 MN case) is a consequenceof the higher wireless load of the different enhance-ments. A higher number of data packets sent throughthe wireless channel and signaling load yields a longerchannel access delay and higher collision rate, result-ing in a higher number of packets waiting to be sent inthe current MN’s AR interface queue at the end of thesimulation and, therefore, lower bandwidth achieved.As commented above, H+F MIPv6 and HMIPv6 ex-perience a lower load on the wireless channel sincemost of the packets are lost in the HA.

For the following studies we have focused on thecase of 20 MNs since this represents the case with ahighest number of MNs in the network where the can-nel can still be accessed without experiencing a highdegradation in the quality of service due to competingnodes.

V.B. Impact of handoff rate

In Section V.A we have shown some of the perfor-mance improvements obtained introducing the MIPv6enhancements. However, as explained in Section IIseveral additional signaling messages have been in-troduced to achieve those results. A trade-off betweenadditional signaling load and performance improve-ment has to be considered. In Figure 9 we study thedifferences in signaling load between MIPv6 and theproposed enhancements for a handoff rate range vary-ing from 0 to 10 handovers per minute for a simulationof 125 seconds.

H+F MIPv6 presents the higher signaling loadwithin the local domain, as expected, since it intro-duces the HMIPv6 signaling load plus the FMIPv6signaling load. The next highest signaling load withinthe local domain belongs to FMIPv6 since, in theevent of a handoff, a higher number of signaling mes-sages are required. One of the purposes of HMIPv6 isto keep constant the signaling load outside of the localdomain. Figure 9 shows that this goal is achieved byHMIPv6 and H+F MIPv6. In the scenarios where aMAP is placed on the CR and when roaming withinthe local domain, HA and CNs do not realize anychange in the point of attachment and receive onlyperiodic BUs, therefore the signaling load is constantoutside the local domain. However, with standardMIPv6 and FMIPv6, when a MN performs a hand-off, it must inmediately inform its HA and CNs, andthus, although the periodic BUs are re-scheduled, thetotal signaling load is increased within and outside thelocal domain. Note though, that the introduction of aMAP in the system results in a quantitative increaseof the signaling load in the local domain, i.e., addi-tional MAP’s BU-BAck plus the encapsulation for theBAcks originated by the HA.

As we can observe, MIPv6 and FMIPv6 introducethe same signaling load outside the local domain sinceall the additional FMIPv6 signaling is sent only withinthe local domain. The same case applies to H+FMIPv6 and HMIPv6 that only differ in the signalingbehavior within the local domain obtaining thus, thesame results outside the local domain.

The signaling load corresponding to standardMIPv6 presents, a priori, a strange behavior having alocal minimum for the case of 8 handoffs/min. How-ever, if we recall that for each handoff the MN re-schedules the periodic BUs to be sent we realize thatif the timer of the periodic BUs is below the time be-tween two consecutive handoffs we will observe theperiodic BUs and afterwards the ones due to a hand-off. On the other hand, if the time between two con-secutive handoffs is below the timer of the periodicBUs, they will be always re-scheduled without beingsent during the whole simulation. Thus, in the case of8 handoffs/min, considering a timer of 10 seconds forthe periodic BUs, they are always re-scheduled due toa handoff and never sent, resulting in a reduction ofsignaling load compared to the previous case.

V.C. Impact of number of correspondent nodes

One of the advantages of HMIPv6 is that when per-forming a local handoff the only entity that has tobe informed via a BU is the MAP, which reduces


0

200

400

600

800

1000

1200

1400

1600

0 2 4 6 8 10

Sig

nalin

g Lo

ad (

bits

/s)

Number of Handoffs/min

H+F MIPv6 within local domainFMIPv6 within local domainHMIPv6 within local domainMIPv6FMIPv6 outside local domainHMIPv6 outside local domainH+F MIPv6 outside local domain

Figure 9: Impact of handoff rate on signaling load

the signaling load. This becomes specially importantwhen the number of correspondent nodes increases,i.e., while the number of BUs to be sent increase lin-early with MIPv6 and FMIPv6 remain constant forHMIPv6 and H+F MIPv6. However, HMIPv6 andH+F MIPv6 do not reduce the number of periodic BUsto be sent but increase it by the additional one sent tothe MAP. Based on the above comments, a trade-offhas to be considered between the number of handoffsperformed within periodic BU periods and the numberof correspondent nodes. This trade-off was alreadyaddressed in [8].

Figure 9 shows the impact of increasing the num-ber of correspondent nodes over the signaling load forthe different protocols in the case of a mobile nodeperforming 4 handoffs4 in 120 seconds. FMIPv6 andH+F MIPv6 perform exactly as MIPv6 and HMIPv6,respectively, concerning to the signaling load sent out-side of the local domain since there are no differ-ences in the protocol behavior for the signaling mes-sages sent outside of it. As we can observe, the us-age of HMIPv6 or H+F MIPv6 reduces the signal-ing load outside of the local domain compared toMIPv6 and the difference tends to increase accordingto larger number of correspondent nodes. The differ-ence though, is not very big since in our scenario thenumber of handoffs per periodic BU periods is smallresulting in a small differentiation of HMIPv6. For ascenario with higher mobility or with larger BU pe-riods the HMIPv6 signaling load reduction would belarger including also the local domain signaling.

4In [10] a twelve-week trace of a building-wide local-areawireless network was studied. The results presented there showedthat 2 handoffs per minute is a high handoff rate for pedestrianmobile users

0

200

400

600

800

1000

1200

1 2 3 4 5 6

Sig

nalin

g Lo

ad (

bits

/s)

Number of Correspondent Nodes

H+F MIPv6 within local domainFMIPv6 within local domainHMIPv6 within local domainMIPv6FMIPv6 outside local domainHMIPv6 outside local domainH+F MIPv6 outside local domain

Figure 10: Impact of number of correspondent nodeson signaling load

V.D. Impact of wired link delay

We have measured the differences in handoff latencyand packet losses between MIPv6 and its enhance-ments when the wired link delayld from the CR tothe HA and CN is increased. The differentld valuesmodel different ‘distances’ to the HA and CNs.

MIPv6’s enhancements reduce the time that elapsesbetween a MN change of point of attachment and thetraffic redirection to its nCoA by introducing a newforwarding entity within the local domain, either oARor MAP, responsible to re-direct the traffic. Thus, thedelay experienced by the re-directed traffic does notdepend on ‘how far’ is the MN from its HA and CNsoutside of the local domain. On the other hand, withMIPv6, the BUs sent after performing the handover,have to reach the HA and CNs (outside of the localdomain) in order to send the traffic to the proper CoAresulting on a direct dependence with theld value.

As we can see in Figure 11 the results are as ex-pected: while an increase in the wired link delay im-plies an increase in the handoff latency for MIPv6, itdoes not affect the other proposals’ handoff latency.

V.E. Impact of random movement

Mobile users are unaware of overlapping areas wherehandoff decisions are taken. This section studieswhether the differences on the performance metricsobserved in previous sections for a mobile node fol-lowing a deterministic path still hold considering ran-dom movement. Note that unexpected movements canhave a quite negative effect on the packet losses expe-rienced due to back and forth movements around theoverlapping areas. This effect could potentially pre-vail over the protocol enhancements.

Figure 12 shows the histogram of packet losses ex-perienced by the studied mobile node moving ran-


0

0.2

0.4

0.6

0.8

1

50 100 150 200 250 300 350 400

Late

ncy

(s)

Wired Link Delay, ld(ms)

MIPv6HMIPv6FMIPv6

H+F MIPv6

Figure 11: Impact of wired link delay on handoff la-tency

0

0.2

0.4

0.6

0.8

1

FMIPv6

H+F MIPv6

HMIPv6

MIPv6

1 10 100 >100

Packet Losses

Num

ber

of occ

urr

ence

s (%

)

Figure 12: Packet losses histogram considering ran-dom movement

domly in the case of 20 mobile nodes for the four dif-ferent protocols. The packet losses occurrences havebeen grouped in lower or equal than 1, 10, 100 andover 100. As we can observe from the figure, the re-sults are consistent with the ones presented in SectionV.A. FMIPv6 and H+F MIPv6 show the better packetloss performance keeping for most of the cases valuesbelow or equal to 1. HMIPv6 outperforms MIPv6 butwithout reaching the level of the protocols that includethe fast handover approach.

V.F. Impact of traffic sources

Until this section we have studied the impact of dif-ferent parameters over a target station receiving a highconstant traffic load (probe) in order to obtain resultswith a significant precision and without the interfer-ence of source burstiness (VoIP, Video) or recoverymechanisms (TCP). In this section we repeat the ex-periment of Section V.A but considering more realis-tic traffic sources and a simulation scenario where all

the MNs send or receive the same type of traffic atthe same rate. By doing this, we analyze whether thedifferent performance improvements observed in pre-vious sections are affected by the traffic source type,i.e., whether a user would realize a service improve-ment or the improvements are ‘masked’ by the traf-fic sources characteristics. Specifically, three differ-ent types of traffic are studied: VoIP, video and TCPtransfers.

As explained in Section III, our VoIP source pro-duces bursty traffic following an on-off Markov pro-cess that results in a high variance between packet ar-rivals. Figure 13 shows the impact of the number ofstations over the packet loss rate of VoIP traffic untilthe congestion level is reached. Since VoIP sourcesproduce a relatively low traffic load (' 24kbps persource) no packet loss is observed for any of the pro-tocols until the 20 MNs case. In this case, surpris-ingly, MIPv6 is the protocol that performs best inpacket losses terms and HMIPv6 worst. The addi-tional load introduced by the different enhancementsin the wireless channel is the reason for this behav-ior. HMIPv6 is the worst one due to the encapsu-lation of all packets directed to the MNs performedby the MAP, FMIPv6 performs better since is ‘betterequipped’ to avoid packet losses and H+F MIPv6 isin the middle since is the one producing more over-head but equipped as well with a mechanism to re-duce packet losses. When the wireless channel is con-gested, i.e., 30 MNs case, we observe the same behav-ior as the one already described in Section V.A. Wecan conclude that, for a scenario with low rate trafficsources sending small packets (compared to the ad-ditional encapsulation header) and in no congestionconditions, the overhead introduced by the differentenhancements would result in a worse performance inhandoff latency an packet losses terms compared tothe baseline Mobile IPv6.

The H.263 video source produces packets of differ-ent length at a variable bit rate for a target rate of 64kbps. We show the impact of the number of stationsover the handoff latency. As we can observe in Figure14, the results are similar to the ones already describedin Section V.A, i.e., H+F MIPv6 and FMIPv6 are theones that perform best in handoff latency terms andMIPv6 is the worst. In this case, in contrast to theVoIP one, the implementation of the Mobile IPv6 en-hancements results, as expected, in a better user expe-rienced service since the additional signaling load isless relevant compared to the data traffic load.

Finally, we study whether a regular user download-ing a file using TCP would notice any difference in


0

100

200

300

400

500

600

700

800

5 10 15 20 25 30

VoI

P P

acke

t Los

ses

Number of Stations

FMIPv6H+F MIPv6

HMIPv6MIPv6

05

101520253035

2 4 6 8 10 12 14 16 18 20

Figure 13: Impact of number of sources on VoIPpacket losses by a receiving user

0

1

2

3

4

5

6

7

8

5 10 15 20 25 30

Vid

eo H

ando

ff La

tenc

y (s

)

Number of Stations

FMIPv6H+F MIPv6

HMIPv6MIPv6

00.050.1

0.150.2

0.250.3

0.350.4

2 4 6 8 10 12 14 16 18 20

Figure 14: Impact of number of sources on Videohandoff latency by a receiving user

the received service by using one of the different pro-posals. For a user performing a download, handofflatency or packet loss rate are not relevant perfor-mance metrics but the experienced bandwidth duringthe TCP transfer is of major interest.

Figure 15 shows the differences on the availablebandwidth for TCP users depending on the MIPv6protocol enhancement used. In the figure we can ob-serve the TCP sources adjustment of the sending rateto the available channel capacity when the number ofmobile users increases. For a number of mobile nodesbelow 10, a lower packet loss rate obtained via theenhancements results in users achievement of largerbandwidth. H+F MIPv6 presents better packet lossesresults than FMIPv6; however, with the latter proposala larger bandwidth value is obtained. In Section V.Awe have shown that there is not a direct relationshipbetween packet losses experienced and obtained band-width. The reason is that in the H+F MIPv6 case theMAP encapsulates all the data packets addressed tothe mobile nodes, and this overhead reduces the avail-able bandwidth in the wireless channel. The same ex-

0

100

200

300

400

500

600

700

5 10 15 20 25 30

TC

P B

andw

idth

(kb

ps)

Number of Stations

H+F MIPv6HMIPv6

MIPv6FMIPv6

0

20

40

60

80

100

10 15 20 25 30

Figure 15: Impact of number of sources on TCP band-width obtained by a receiving user

planation applies to HMIPv6, where the lower packetloss rate does not result in a significant higher band-width compared to MIPv6 because of the packet en-capsulation within the local domain.

When the number of mobile nodes increases, theprobability of experiencing a collision while trying toaccess the channel increases, too. This, in turn, trig-gers the TCP congestion avoidance mechanism moreoften reducing the packet losses experienced by theMNs and thus, decreasing the bandwidth differencesbetween the proposals. These differences would oth-erwise be much bigger, as it has been shown is SectionV.A, when the users try to get a larger bandwidth thanthe one actually available in the channel.

As a conclusion, TCP users would also benefit fromthe implementation of one of the MIPv6 protocol en-hancements even though the improvement would belower than for other types of traffic, e.g., CBR.

VI. Conclusion

Mobile IPv6 represents a key element of futureAll-IP wireless networks to allow a user to freely roambetween different wireless systems. In this paper wehave provided quantitative results on Mobile IPv6 per-formance as experienced by a mobile node and on thelevel of improvement that can be achieved by usingthe proposed Mobile IPv6 enhancements. The resultswere achieved through a thorough study via simula-tion that required to implement Neighbor Discovery,HMIPv6, FMIPv6 and our combination of HMIPv6and FMIPv6 forns-2.

We performed a ‘stress test’ of the protocols wherewe studied how handoff latency, packet loss rate, ob-tained bandwidth and fast handoff process successprobability are affected by the number of mobilenodes, i.e., by competition for the wireless medium,


or by protocol interactions, e.g., with the NeighborDiscovery process of IPv6. The behavior of the pro-tocols for a general case considering random move-ments and more realistic traffic sources, i.e., VoIP,video and TCP, were also studied. Finally, the sig-naling load costs associated to the different proposalscompared to the performance improvements obtainedwere analyzed, considering a broad range of handoffrates and number of correspondent nodes. These fac-tors were shown to have a significant influence overthe performance metrics and we indicated the pointsto be taken into account in a real implementation.

Specifically, we have shown that while some sim-ulation results corroborate the intention of the proto-cols specifications, other results give insights not eas-ily gained without performing simulations. Some ofthe key results are thati) random movements of theobserved mobile node do affect the experienced per-formance but the improvements with respect to theperceived quality of service when using one of the var-ious protocol enhancements is still clearly noticeable,ii) in scenarios where the users produce a low rate withsmall packets, e.g, VoIP sources, the additional over-head introduced by the proposed enhancements canresult in a worse performance than the baseline Mo-bile IPv6 one, andiii) Mobile IPv6 can eventually out-perform its proposed enhancements in packet lossesterms in saturation conditions due to the higher num-ber of packets discarded directly that lower the load inthe wireless channel.

Through this analysis a deep insight on the differ-ent overall performance of the various protocols andtheir causes was acquired. Therefore, the results ofthis study are twofold. First, we provided quantitativeresults for the different IETF proposals of the overallperformance for a realistic ‘hot spot’ scenario. Sec-ond, we provided the reasoning behind the impact ofthe different parameters over the performance of thevarious protocols in saturation and no saturation con-ditions. This reasoning can applied when other sce-narios are analyzed.

VII. Acknowledgments

This work has been partially supported by the ISTproject Moby Dick [6]. The authors would like tothank Albert Banchs and the anonymous reviewers fortheir helpful comments.

References

[1] Mobiwan: ns-2 extensions to study mo-bility in Wide-Area IPv6 Networks.

http://www.inrialpes.fr/planete/mobiwan.

[2] Network Simulator (ns), version 2.http://www.isi.edu/nsnam/ns.

[3] Artificial conversational speech. ITU-T p.59,Recommendation, 1993.

[4] H.263, video coding for low bitrate communica-tion. ITU-T/SG15 Recommendation, 1996.

[5] Wireless LAN Medium Access Control (MAC)and Physical Layer (PHY) Specifications. IEEEStandard 802.11, June 1999.

[6] Mobility and Differentiated Services in a FutureIP Network. IST-2000-25394, 2000.

[7] C.Bettstetter, H.Hartenstein, and X.Perez-Costa.Stochastic Properties of the Random WaypointMobility Model. to appear in ACM KluwerWireless Networks, special issue on Modeling &Analysis of Mobile Networks (WINET), 2003.

[8] C.Castelluccia. HMIPv6: A Hierarchical MobileIPv6 Proposal. ACM Mobile Computing andCommunication Review (MC2R), April 2000.

[9] C.Perkins. IP encapsulation within IP. RFC2003, October 1996.

[10] D.Tang and M.Baker. Analysis of a Local-AreaWireless Network. ACM International Con-ference on Mobile Computing and Networking(MOBICOM), 2000.

[11] C. (Ed.). IP Mobility Support for IPv4. RFC3344, August 2002.

[12] A. C. et al. Comparison of IP Micro-mobilityProtocols. IEEE Wireless CommunicationsMagazine, Vol. 9, February 2002.

[13] C.-N. et al. QoS Provisioning Using a ClearingHouse Architecture. International Workshop onQuality of Service (IWQoS), June 2000.

[14] H. J. et al. Fast Handover for HierarchicalMIPv6 (f-hmipv6). Internet Draft, work inprogress, June 2003.

[15] R. et al. Performance analysis on HierarchicalMobile IPv6 with Fast-handoff over End-to-EndTCP. IEEE Global Telecommunications Confer-ence (GLOBECOM), 2002.

[16] T. et al. Neighbor Discovery for IP Version 6.RFC 2461, December 1998.


[17] W. et al. Mathematical Statistics with Applica-tions, p.368. 4th Ed.PWS-KENT, 1990.

[18] F. Fitzek and M.Reisslein. MPEG-4 and H.263Video Traces for Network Performance Evalua-tion. IEEE Network,Vol. 15,No. 6,pages 40-54,November/December 2001.

[19] G.Tsirtsis, A.Yegin, C.Perkins, G.Dommety,K.El-Malki, and M.Khalil. Fast Handovers forMobile IPv6. Internet Draft, work in progress,March 2003.

[20] H.Soliman, C.Castelluccia, K.El-Malki, andL.Bellier. Hierarchical MIPv6 mobility man-agement (HMIPv6). Internet Draft, work inprogress, June 2003.

[21] D. Johnson, C.Perkins, and J. A. (Eds.). Mo-bility Support in IPv6. Internet Draft, work inprogress, June 2003.

[22] M.Torrent-Moreno, X.Perez-Costa, andS.Sallent-Ribes. A Performance Study ofFast Handovers for Mobile IPv6. in Proceedingsof IEEE Local Computer Networks (LCN),September 2003.

[23] N.Montavont and T.Noel. Handover Manage-ment for Mobile Nodes in IPv6 Networks. IEEEWireless Communications Magazine, August2002.

[24] S. Thomson and T.Narten. IPv6 Stateless Ad-dress Autoconfiguration. RFC 2462, December1998.

[25] X.Perez-Costa and H.Hartenstein. A Simula-tion Study on the Performance of Mobile IPv6in a WLAN-Based Cellular Network. ComputerNetworks, special issue on ‘Towards a New In-ternet Architecture’, September 2002.

[26] X.Perez-Costa and M.Torrent-Moreno. A Per-formance Study of Hierarchical Mobile IPv6from a System Perspective. in Proceedings ofIEEE International Conference on Communica-tions (ICC), May 2003.

[27] X.Perez-Costa, M.Torrent-Moreno, andH.Hartenstein. A Simulation Study on thePerformance of Hierarchical Mobile IPv6.in Proceedings of International TeletrafficCongress (ITC), 2003.

VIII. Biographies

Xavier Perez Costareceived the M.Sc. degree inelectrical engineering from the Polytechnic Universityof Catalonia (UPC) in 2000. He did his master thesisat NEC Network Laboratories Europe in Heidelberg(Germany) in the area of QoS provisioning in IEEE802.11 wireless LANs. After the completion of hisdegree he was hired by NEC and focused on IP-basedmobility management issues in the framework of theEuropean projectMoby Dick. Currently he is a Re-search Staff Member at the 3G Technologies Groupworking in the evolution of UMTS networks towardAll-IP networks and a Ph.D candidate at the Telem-atics Engineering department of UPC. His researchinterests include among others, wireless communica-tions, mobile networking, and quality of service.

Marc Torrent Moreno received his degree inTelecommunications Engineering in January 2003from the Polytechnic University of Catalonia. Prior tograduation he worked at British Telecom (UK, 2001)and at NEC Network Laboratories Europe (Germany,2002) performing research in mobile IP networks forhis graduation project. In February 2003 he startedworking as a research assistant at DaimlerChryslerResearch and Technology North America and focusedon the design and development of WAVE technologyin vehicle-to-vehicle communications. Since January2004 he is a Ph.D. candidate at the Telematics Insti-tute of the University of Karlsruhe, Germany. Hiscurrent research interests include wireless communi-cations, ad-hoc networks and network security.

Hannes Hartenstein is a professor at the Univer-sity of Karlsruhe, Germany, affiliated with the Insti-tute of Telematics and the University’s ComputingCenter. He received the diploma degree in mathe-matics in 1995 and the Ph.D. degree in computer sci-ence in 1998, both from Albert-Ludwigs- Universitat,Freiburg, Germany. He was Erasmus Scholar withthe University of East Anlia, Norwich, U.K., in 1991-1992 and received the Capocelli Award from the IEEEData Compression Conference 1997 (with M. Ruhl)for the paper ”Optimal fractal coding is NP-hard.” Hejoined NEC Network Laboratories Europe in 1999 asa member of the Mobile Internet group. He currentlyfocuses on IP-based mobility management as well ason ad hoc routing. His general research interests in-clude mobile communication, networking, multime-dia, image/video processing, and theoretical computerscience.


Date post:	09-Oct-2014
Category:	Documents
Upload:	amine-tounsi
View:	19 times
Download:	0 times

COMP NW - Performance MIPv6 FMIPv6 HMIPv6 and Their Combination-preprint

Documents