Target access router selection in advanced mobility scenarios

D. Di Sorte*, M. Femminella, L. Piacentini, G. Reali

Dipartimento di Ingegneria Elettronica e dell’Informazione (D.I.E.I.), University of Perugia, via G. Duranti 93, 06125 Perugia, Italy

Received 1 April 2004; revised 27 April 2005; accepted 9 May 2005

Available online 2 June 2005

Abstract

To provide nomadic users with QoS-enabled services, advanced mobility management will prove to be of fundamental importance in the

future Internet. Mobile IP is the reference protocol supporting mobility at IP level. However, it has been widely recognised that it could

perform poorly, especially with QoS demanding applications. In order to improve Mobile IP performance, the discovery and selection of the

target access router to hand over to will play a crucial role. In this paper, we present a novel approach for Candidate Access Router Discovery

and Target Access Router selection. The process is fully distributed, multicast-based, and allows timely intra and inter-access router

handover, which can be either intra or inter-technology. The effectiveness of the solution to drive mobile terminal handovers has been tested

by simulations.

q 2005 Elsevier B.V. All rights reserved.

Keywords: IP mobility; Multicast; Candidate access router discovery; Access router selection

1. Introduction

IP mobility protocols (i.e. Mobile IPv4/v6, [5,6]) enable

mobile nodes (MNs) to execute IP layer handover between

access routers (ARs). Basic Mobile IP (MIP) protocols are

well known for their poor performance and inability to

support a seamless handover. In [1], the Authors define the

seamless handover as ‘a handover in which there is no

change in service capability, security, and quality’.

A number of different approaches to improve MIP have

been proposed so far: (i) micro-mobility solutions [7], which

aim to limit the handover range of MIP procedures, thus

reducing handover latency and signalling burden; (ii)

context transfer solutions [3,4], the goal of which is to

quickly re-establish information states (context) associated

with the MN in the new AR upon handover, thus avoiding

re-initiation of the signalling procedure to set-up the service

from scratch; (iii) fast handover procedures to minimise

packet losses and latency [10,11].

0140-3664/$ - see front matter q 2005 Elsevier B.V. All rights reserved.

doi:10.1016/j.comcom.2005.05.004

* Corresponding author. Tel: C39 075 585 3630; fax: C39 075 585

3654.

E-mail addresses: disorte@diei.unipg.it (D. Di Sorte), femminella@

diei.unipg.it (M. Femminella), piacentini@diei.unipg.it (L. Piacentini),

reali@diei.unipg.it (G. Reali).

All the proposed enhancements assume that the new AR

to hand over to is known. This is indeed the missing step in

the overall seamless handover procedure. Thus, in advanced

mobility scenarios, it is important to discover the set of

potential target ARs and to select the most appropriate one

before handing over. A candidate access router discovery

(CARD) procedure collects information about the ARs

which are candidates (CARs) for the MN handover [2,8,9].

It is then possible to identify the Target AR (TAR), i.e. the

one that best matches the MN’s requirements and the CARs’

capabilities.

The CARD requires two essential functions to be

performed:

† reverse address translation, which means to use layer 2

addresses to determine the coverage areas which is

served through a port of an AR, thus with its IP address

(layer 3 coverage). This is important to speed-up MIP

handovers, and is implemented by binding the layer 2

identifier (L2 ID) of each access point (AP) with the IP

address of the CAR connected to it. MNs are assumed to

listen to the L2 beacons transmitted by surrounding APs

periodically and to learn their L2 IDs. This information

is then exchanged with the current AR. The CARD

procedures can then support MNs with layer 3

information. Thus, in order to start the handover process,

MNs do not have to wait for the MIP router

Computer Communications 29 (2006) 337–357

www.elsevier.com/locate/comcom

D. Di Sorte et al. / Computer Communications 29 (2006) 337–357338

advertisement from the new AR and fast handover

procedures may be executed ([10,11]). Note that the

capacity of performing frequency scanning is actually a

minimum requirement for all wireless technologies;

† discovery and update of service capabilities (SCs) of

CARs. In general, an SC is the set of parameters

(available bandwidth, price, security support, radio

access technology, etc.) that characterises the network

service provided to the MN. An SC includes all the most

important parameters that influence handover decisions.

Thus, the final aim is to drive the MN operation

according to specific objectives, such as (i) to balance

the traffic load on the access network, (ii) to drive the

MN towards the cheapest wireless access, (iii) to drive

the MN towards an access which guarantees the same

level of quality of service (QoS).

The first goal of this paper is to present a novel CARD

approach.

We start by assuming that the network is controlled by a

single operator (Fig. 1 shows the reference network

scenario) and that the provider of the IP connectivity

(Connection Service Provider, CSP) is also the owner of the

network infrastructure (Network Service Provider, NSP).

The result of this assumption is that, without any

commercial agreement there are no logical reasons for a

CSP to induce its mobile customers to move to other access

domains. In fact, this would imply a loss of traffic and

revenues in favour of competitors. Moreover, unless specific

agreement has been made, any CSP is usually unwilling to

give other organisations confidential information regarding

its own network (current bandwidth, security policies, etc).

Thus, we assume that our CARD procedure will be confined

Fig. 1. Reference net

within a single administrative domain. Then, we also

indicate an extension to manage the case of different

NSPs co-ordinated by a single CSP.

The proposed Push-Mode-Multicast CARD (PMM

CARD), already outlined in [22], is fully distributed and

allows each AR to dynamically self-construct a map of the

surrounding wireless coverage; the main innovation of our

approach with respect to previous proposals is the use of

push-mode multicast transmissions. We use multicast in a

push-mode in order not only to strongly reduce the latency

due to explicit queries, but also to lower the signalling

burden. The advantages in terms of signalling burden and

discovery time of the PMM CARD solutions with respect to

those proposed by the IETF within the Seamoby WG [2] are

presented in [22].

In this work, we assume that the access network is

heterogeneous and that terminals may have reconfiguration

capabilities so as to be able to access different radio access

technologies (RATs). We assume that all terminals have a

single active radio interface at a given time to save power.

This also implies the constraint of MNs not being allowed to

scan for beacons of RATs different from the one currently

being used, with the result that they cannot obtain coverage

information from them.

In homogeneous networks, handover is typically driven

by metrics which are strictly related to the received power

level [17]. In heterogeneous mobile environments, more

complex metrics combining a higher number of parameters

have to be defined (e.g. price, bandwidth, priority, power

consumption, reliability, etc.) [18–20].

In this paper, we present a novel TAR metric capable of

driving layer 2 and layer 3 handovers. Our goal is to support

intra and inter-AR handovers, which can be intra and

work scenario.

D. Di Sorte et al. / Computer Communications 29 (2006) 337–357 339

inter-technology. The metric takes into account the

bandwidth availability and received power level, together

with a statistical factor which helps to drive inter-

technology handovers by inferring about layer 2 coverage.

The paper is structured as follows. In the Section 2, we

summarise the current state of the art about CARD. In

Section 3, our PMM CARD procedure is described in detail.

In Section 4, the TAR approach is presented together with

the selection metric used. In Section 5, we describe the

simulation model and analyse the performance of the

procedure obtained from simulation trials. In particular, we

show results concerning the capability of the CARD/TAR

approach to drive inter-technology handovers. Finally,

Section 6 reports some concluding remarks.

2. Related works

In order to enable a seamless IP-layer handover, a CARD

solution must include two important tasks: (i) the reverse

address translation from L2 IDs; (ii) the discovery and

update of SCs. As mentioned the introduction, the former

function is needed to speed up the handover process,

whereas the latter function provides the TAR selection

process with the necessary information about the service

capabilities of the candidate wireless accesses in order to

drive the MN towards the most appropriate AP and AR. The

scope of the CARD process is an IP network with a number

of ARs that control a set of APs, which may be

heterogeneous. Handovers can be both inter and intra-

technology, intra and inter AR, at both layer 2 and layer 3.

We assume that each AR manages a number of APs;

consequently, the coverage area of an AR is the union of the

coverage areas of the relevant APs. Two ARs are

neighbours if their wireless coverage areas overlap.

Since MNs are generally assumed to be able to

periodically listen to the L2 beacons transmitted by the

surrounding APs, in [21] the authors propose an initial,

straightforward solution to enable the two CARD functions.

This consists of embedding the IP address, IP prefix, and

service capabilities in L2 beacons. However, this approach

has three main drawbacks. The first is strictly conceptual:

encapsulating layer 3 data within layer 2 control frames

violates the protocol layer architecture. The second is that

Fig. 2. Steady phase: message exchange for a

this approach would require some modifications to the

standard of existing technologies. Finally, to save wireless

bandwidth, it is not recommended to stuff beacons with a

large amount of additional information.

To overcome the aforementioned drawbacks, it is

necessary to define a network-assisted CARD process.

The solution presented in the framework of the IETF

Seamoby WG is as follows. When an MN remains under the

coverage of an AR (current AR), it obtains L2 IDs from the

beacons of some other APs, and passes them to its current

AR at a TAR event. The current AR is in charge of

providing the MN with either the IP addresses of CARs and

the relevant service capabilities (if the TAR algorithm is run

in the MN), or the TAR IP address (if the selection

algorithm is carried out at the current AR). This means that

not only the signalling messages to resolve L2 addresses

between an MN and its current AR, but also service

capabilities must be exchanged. Thus, a basic requirement is

that each AR is able to obtain information about the state of

the neighbouring wireless coverage (i.e. the AR–AP pairs)

and store it in a local cache. The state stored by the AR

should contain the IP address of the neighbouring ARs, the

L2 IDs of the relevant APs and the service capabilities

associated with the AR–AP pairs. This information may be

used by the current AR to offer an alternative wireless

access to an MN under its own coverage. The operations

described need a continuous, dynamic signalling exchange

between the network entities involved in the CARD process.

The complete message exchange between the MN, its

current AR, and a CAR is depicted in Fig. 2. This exchange

takes place during the ‘steady phase’ of the CARD process

on completion of the coverage map in the current AR. The

capability exchange between the current AR and CARs may

be performed either upon an MN request, or when the

capability lifetime expires. In Fig. 2, we have assumed that

the MN is in charge of both triggering and performing the

TAR selection. The MN communicates the current list of L2

IDs received (step 2), since they are candidate wireless

accesses for a possible handover, and must be considered for

the TAR selection.

As regards the message exchange in the steady phase,

some security considerations apply. Since the SC exchange

is the key element of any CARD process, IPsec Security

Associations (SAs) [23] among ARs is a valuable tool to

ddress mapping and capability update.

D. Di Sorte et al. / Computer Communications 29 (2006) 337–357340

protect it. Integrity check and authentication are mandatory

IPsec functions, whereas ensuring confidentiality is up to the

specific operator’s policy. As regards message exchange

between an MN and an AR, an SA is definitely needed for

an integrity check and authentication when the AR sends

CARD/TAR information to the MN, since false information

would drive the MN towards an incorrect, or bogus TAR. In

addition, if an operator wishes to hide information

exchanged on the air interface concerning network

capabilities, confidentiality also has to be supported.

An important issue of any CARD approach is the

discovery of the IP addresses of the neighbouring ARs

(‘discovery phase’), and the determination of the (L2 ID)/(IP address) mapping to be stored in the local cache.

Specifically, the discovery phase consists of discovering the

neighbouring ARs/APs, so as to build a map of the wireless

coverage of the surrounding area at both layer 2 and layer 3

for each AR. This information is clearly useful only in the

case of dense wireless coverage. This situation is realistic

for an operator offering a seamless wireless network service,

which can deploy a CARD mechanism to support TAR.

The accuracy of a CARD solution greatly depends on the

ability of an AR to obtain accurate information about

geographically neighbouring ARs, and to exchange SCs

with them. A further consideration is that a manual, address-

mapping configuration at ARs would not be a good solution

because (i) it is not feasible in large wireless networks; (ii) it

would not permit any adaptation to variable network

topologies; (iii) the coverage area of an AR (i.e. of its

APs) could not be easily and exactly determined. For these

reasons, a solution for automatically building the coverage

map at each AR is needed. Below we illustrate the solutions

proposed in literature, which can be classified in two

separate, general schemes: (i) handover-based solutions,

which use the first plain MIP handover between two ARs to

infer knowledge about the surrounding wireless coverage;

(ii) solutions which analyse and estimate the radio coverage

situation by using the information obtained by MNs from L2

beacons.

2.1. Handover-based CARD approach

The basic idea of a handover-based solution [2,9] is that

two ARs discover they are neighbours once an MN has

executed a plain MIP handover between them. Thus, an AP

connected with a neighbouring AR is discovered after a

plain MIP handover is accomplished through that particular

AP, which implies a layer 2 handover. After this handover,

the MN has to send a signalling message to the new, current

AR containing the IP address of the old AR and the L2

address of the old AP. Thus, if the association (L2 ID)/(IP

address) was unknown, the current AR can create a new

entry in its cache. Then, the new current AR is in charge of

informing the old one by means of a specific signalling

message that both of them are neighbours, so that the old AR

can update its own coverage map as well.

This message exchange can be used by the new AR to

check with the old AR to see whether the MN was indeed

attached to it during a reasonable past interval so as to trust

its information.

The handover-based approach presents some drawbacks.

Firstly, the first handover, involving an as yet undiscovered

AP by the old AR towards an AR (bootstrap handover),

cannot be driven by the CARD procedure; this could be a

weakness not only for delay-sensitive applications, but also

in a highly dynamic access network topology. In addition,

the time needed to complete the discovery phase could

increase in the case of dense wireless coverage. In fact, in

the latter case, handovers are driven by the TAR process

towards previously discovered APs, so that a non-driven

handover (discovery event) is carried out only when there is

no alternative. Finally, if an MN switches between two APs,

whose coverage areas are very close but do not overlap, then

this would result in false coverage information sent to the

network. In other words, it would imply an enlargement of

the CARD table with useless coverage data.

2.2. Centralised L2 beacon-based approach

This solution is illustrated in [2,8]. The basic idea is that

each AR must register with a centralised server, by

indicating its own IP address and the L2 IDs of the APs

associated with it. This server is a database that is

dynamically updated by ARs. Its task is to process queries

from ARs to resolve L2 IDs and to contribute to the building

of the wireless coverage map in ARs.

The process develops as follows. When an MN discovers

the L2 IDs of APs during frequency scanning, the MN

passes one or more of them to its current AR. If unknown, the

current AR sends them to the centralised server, which

returns the IP addresses of the relevant ARs. This mechanism

enables those (AR,AP) pairs with wireless coverage

overlapping with the one of the current (AR,AP) pair to be

identified. Thus, MNs can be regarded as L2 sensors used to

identify candidate wireless accesses to hand over to.

We expect an L2 beacon-based solution to speed up the

discovery phase when compared with the handover-based

approach [22], due to the higher number of events (L2

beacon listening), which trigger the discovery of new

wireless resources.

As regards security considerations, this message

exchange must establish specific SAs to guarantee integrity,

authentication and, if needed, confidentiality. In addition,

malicious MNs could communicate false information (e.g.

L2 IDs of APs that do not have a wireless coverage

overlapping with the current AR). This would cause storing

of incorrect wireless coverage information in routers.

Therefore, as highlighted in [13], if MNs cannot be trusted,

an L2 beacon-based procedure is inefficient from a security

point of view.

The server-based solution proposed by the Seamoby WG

extends the CARD protocol to support an AR-server

D. Di Sorte et al. / Computer Communications 29 (2006) 337–357 341

message exchange, whereas in [8], the Authors make use of

the SLP (Service Location Protocol) architecture with a

centralised Directory Agent. ARs act as SLP User Agents,

that is they send service requests to the Directory Agent.

A distributed SLP architecture has also been proposed

in [8].

3. The CARD approach

In this section, we present the Push-Mode-Multicast

CARD (PMM CARD), which is a proposal to perform the

CARD procedure within an administrative domain. Then we

extend the procedure towards an inter-domain scope.

The PMM CARD is network-assisted, distributed and

based on push-mode multicast transmission. Our final goal

is to define a CARD process able to provide the TAR

process with all the necessary information needed to

properly drive intra and inter-technology handovers.

We assume that MNs are either classical single-mode

terminal or adaptive terminals. The latter may be either re-

configurable terminals based on the software defined radio

paradigm or simply multi-mode terminals. In any case, we

assume that they are capable of listening to layer 2 beacons

transmitted by surrounding APs. However, in the case of

multi-mode terminals, we assume that such terminals do not

scan for beacons of RATs other than the one currently in

use. The rationale of this assumption is that, with a number

of network interfaces, scanning for beacons on all interfaces

could result in a large drain on power.

3.1. The discovery phase

As regards the discovery phase, we follow a distributed

L2 beacon-based approach. It allows us (i) to avoid the

drawbacks of a centralised solution, (ii) to avoid the

bootstrap handover, (iii) to improve the precision and

accuracy of the coverage information, and (iv) to speed up

the discovery phase [22].

Each AR organises this information in a specific table

format, the CARD table, where the notation APi,j denotes

the jth AP connected to the ith AR (ARi). An example of a

CARD table within ARh is reported in Table 1, where:

1. the upper rows of the table specify: (a) the L2 IDs and the

RAT identifier of the APs connected to the AR; (b) the

relevant service capabilities; (c) a statistic parameter

Table 1

Example of CARD table stored in ARh

IPAR APL2ID RAT SC P

ARh APh,p 0 SCh,p P

. . . .ARx APx,k 2 SCx,k P

. . . .ARy APyj 1 SCyj P

which provides some sort of coverage information at

layer 2 (e.g. with reference to Table 1, PrAP(h,s)/AP(h,p) is

an estimation of the probability of making a successful

layer 2 handover from APh,s to APh,p). This parameter is

clearly used only if the APs refer to different RATs, since

in this case it is not possible to rely on beacon listening.

In the following subsection we will describe how to

compute such a statistical parameter;

2. the other rows specify: (a) the L2 ID and the RAT

identifier of each geographically adjacent AP; (b) the

corresponding IP address of the relevant ARs; (c) the

associated service capabilities and the parameter report-

ing the probability of successfully completing an

handover towards ARh. It is worth noting that in this

case the probability of success refers to layer 3 handover.

To reduce the complexity of table management, the

entries of the CARD table are soft states, which are deleted

if they are not refreshed within a given time interval. This is

particularly helpful in the case of a dynamic network access

topology.

Part of the CARD procedure is based on the use of a

domain-wide, multicast group MGOP defined by the

network operator. MGOP includes all the ARs that provide

wireless connectivity and act as multicast hosts. MGOP is

used to resolve the IP address of the AR from the L2 ID of

any of its APs. For this purpose, when an AR starts to offer

wireless connectivity via some APs, it has to join MGOP and

to send its IP address to its multicast address, together with

the L2 IDs of the active APs under its IP scope. Thus, all

ARs in MGOP are informed of all the L2 and L3

configuration parameters of the new arrival. In turn, one

of the participants (e.g. the latest AR that has joined MGOP)

sends a unicast reply with the address mapping of all the

ARs currently active in MGOP. Below, this preliminary

phase will be referred to as the ‘initialisation phase’.

Since the coverage area of the new AR does not typically

overlap with the coverage of all ARs, only a subset of the

received information will be used to build the CARD table.

Nevertheless, we remark that the amount of address

mapping data to be exchanged during the initialisation

phase and to be managed within ARs is very limited and

simple.

Active ARs must promptly notify all ARs of all the

variations in their radio coverage (e.g. (de)activation of APs)

via MGOP. Consequently, the interested ARs can quickly

update their CARD table. Moreover, this mechanism speeds

rob APh,s Prob APh,t . Prob APh,n

rAP(h,s)/AP(h,p) PrAP(h,t)/AP(h,p) . PrAP(h,n)/AP(h,p)

. . .rAP(h,s)/AP(x,k) PrAP(h,t)/AP(x,k) . PrAP(h,n)/AP(x,k)

. . .rAP(h,t)/AP(y,j) PrAP(h,t)/AP(y,j) . PrAP(h,n)/AP(y,j)

Fig. 3. Message exchanges when a new AP is detected by an AR by means

of beacon listening by an MN.

D. Di Sorte et al. / Computer Communications 29 (2006) 337–357342

up the address resolution phase, thus avoiding the latency of

consulting a remote entity. In this sense, the mechanism is

proactive with respect to the time in which the information is

needed. This feature gives a clear advantage over the IETF

solutions, which are pull-mode based.

Once each active AR has stored the complete L2 and L3

address list, the discovery phase may take place. The reader

is invited not to mistake the address list for the CARD table,

which contains the information related to the surrounding

radio coverage at each AR.

Let us assume that an MN, located under coverage of the

sth AP connected to ARh (APh,s), enters the coverage area of

another AP connected to ARz (say, APz,k). The steps of the

discovery phase are (see Fig. 3):

1. the MN listens to the beacons of the new AP;

2. the MN notifies its current AR (ARh) of the L2 ID of the

new AP;

Fig. 4. CARD tab

3. if the detected AP does not appear in any row of its table,

ARh gets the IP address directly from the address list and

asks ARz for the SC relevant to APz,k. In addition, it

invites ARz to join its own local multicast group (namely,

MGh) and sends the SC associated with APh,s;

4. ARz sends a unicast reply, containing the SC associated

with the ARz–APz,k pair and the invitation to ARh to join

MGz;

5. the process ends when ARz and ARh join the multicast

groups MGh and MGz, respectively.

Clearly, the reciprocal invitations to join the local

multicast groups are sent only when two ARs are discovered

to be neighbours, i.e. the first time they discover each other.

As regards the second step, a question arises: when does

the MN send the list of L2 IDs to the current AR? This may

occur either (i) at a TAR event or (ii) periodically, if new L2

IDs have been received during the last Tbeacon seconds. The

latter option would result in a larger use of signalling and

would be particularly useful to better follow dynamic

network topology status.

Step No. 4 is carried out to update the CARD table

maintained in ARh (see Fig. 4), as well as the CARD table

maintained in ARz. If two ARs are discovered to be

neighbours, subsequent discovery events can also create

new entries on the CARD tables relevant to other APs.

To sum up, the procedure enables the geographical

proximity of an AR to be associated with its participation in a

given multicast group. Note that the procedure enables the

coverage map to be extended across all RATs under the

assumption that each RAT is configured as active on a subset

le update.

D. Di Sorte et al. / Computer Communications 29 (2006) 337–357 343

of the terminals moving in the relevant area. In addition, it is

worth noting that an MN can receive beacons (and send them

to the current AR) from all RATs during inter-technology

handover attempts when scanning different RATs.

3.2. The steady phase

After the transient discovery phase, the SC information

stored in CARD tables needs to be maintained. Local

multicast groups are used to manage the geographical

proximity of ARs. The generic ARh, managing the multicast

group MGh, multicasts update messages containing the SCs

associated with its APs. This event may occur both

periodically and as a result of significant variations in the

SCs. These updates are received by all the ARs included in

MGh which are neighbours of ARh. This process enables

ARs to continuously update the relevant SCs of their

neighbours stored in their CARD table. It is important to

remark that updates relevant to a given AR are sent in a

push-mode. In other words, they are controlled by the

sending AR, which decides when additional information

needs to be sent through its local multicast group. This

procedure also enables the signalling burden to be lowered

[22]. The update process can also be considered proactive,

in the sense that any AR always has all information updated

without having to issue an explicit query, as in the IETF

procedure.

Additional considerations regarding the statistical par-

ameters have to be made. Let us consider, for instance,

PrAP(h,s)/AP(x,l); such a value is simply evaluated as the

number of successful handover events between these two

APs, i.e. from APh,s to APx,l, divided by the total number of

attempts. Both the numerator (successful handovers,

NHO,SUCC) and the denominator (total number of handover

attempts, NHO,TOT) are initialised to 1 for each column in the

table stored in ARh, so that the corresponding probability is,

in turn, initialised to 1/1Z1. However, as previously said,

these fields are meaningful only for APs exploiting different

RATs.

In our approach, we consider as SC the amount of

available bandwidth of the link from the wireless interface

of an AP to the output port of the relevant AR. In a single

operator scenario other factors, such as connection price and

security associations, seem to be of less interest than the

available bandwidth, which is the basic parameter to

perform admission control for QoS-enabled services and

to execute load balancing actions.

3.3. Security considerations

As underlined in [15,16], in order to extend MIP to

support advanced network services, additional mechanisms

are needed to authenticate MNs and to authorise connec-

tivity according to local policies (e.g. pricing/accounting

strategies). RFC 2977 [15] describes an infrastructure which

enables AAA servers to authenticate MNs and authorise

them to access the network, whereas, in [16], the Authors

propose some extensions to MIP registration messages to

create Mobility SA between the MN and its home agent,

and/or between the MN and a foreign agent.

Since we propose a procedure to support advanced

network services, we assume that MNs are authenticated

when they access the network. Thus, the coverage

information provided by them and used in the network to

build wireless coverage maps is assumed reliable. False or

inexact coverage information provided by MNs may be due

to malfunction only. This unlucky event could cause an

incorrect expansion of the CARD tables and additional

signalling traffic in the network. Thus, incorrect TAR

selections might be performed, even if only for the MN

providing inexact information about its set of candidate

accesses. However, since the entries of the CARD tables

are soft states, as mentioned above, that incorrect

(and hopefully sporadic) information is not renewed, and

is deleted after a given lifetime.

To support a secure message exchange, an IPsec-based

solution has to be exploited to guarantee authentication,

integrity, and confidentiality. The extension of the

standardised point-to-point IPsec architecture towards

multicast is currently being investigated (e.g. see [14]).

3.4. An extension of the CARD solution towards

an inter-domain scenario

So far, the reference scenario of this paper is

characterised by a single NSP, which is the owner of the

network infrastructure. It is in charge of providing mobile

terminals with IP connectivity, thus acting also as a CSP.

Consequently, the scope of the CARD procedure in such a

scenario is limited to the wireless coverage provided by the

NSP/CSP, since there are no reasons that could lead the

operator to support mobile customers to change the access

domain. In fact, this would clearly mean a loss of traffic

and consequently a loss of revenue in favour of

competitors.

However, in our opinion, there is another typical

network/business model which deserves to be considered.

Such a scenario is characterised by a CSP which, under

specific contractual agreements with a number of different

NSPs, is allowed to use their network infrastructure to

provide its subscribers with wireless connectivity. This

means that the wireless coverage of the CSP spans over a

number of NSPs, and thus an MN currently handled by

that CSP may access Internet services through ARs

belonging to different administrative domains, with

possible overlapping radio coverage. Consequently, the

scope of the CARD process has to be extended to support

handovers among those NSPs which have stipulated the

agreement with the considered CSP, and allows it to use

their network infrastructure to provide a wireless

connection service.

Fig. 5. Inter-domain CARD discovery phase.

1 At this level of the hierarchy, a manual configuration of CGWs may be

envisaged.

D. Di Sorte et al. / Computer Communications 29 (2006) 337–357344

Note that the choice of using multicast transmissions in

our PMM CARD is also inspired by the intra-domain scope

of our CARD approach, since it is well known that inter-

domain multicast transmissions might imply serious

problems. Thus, in this new scenario, the original procedure

needs to be modified to overcome drawbacks due to

multicast transmission through different administrative

domains. A possible solution is briefly described below.

The first step is the introduction of a new network entity

within each NSP, named CARD gateway (CGW), the goal

of which is to manage inter-domain CARD signalling

exchange. We remark that the signalling exchange among

CARD CGWs is managed with unicast transmissions. We

assume that CGW belongs to the MGOP, so that it has

knowledge of address mappings within each NSP built up

during the initialisation phase.

As regards the discovery phase, the basic idea is that the

CGW is polled by an AR only if it has not succeeded in

resolving a specific L2 ID. Then, upon reception of the

query regarding a given L2 ID, the CGW forwards such a

request towards the CGWs of those NSPs that may have

a radio coverage overlapped by the one of the current NSP.1

If a CGW finds the requested mapping in its local address

list, it contacts the AR associated with the L2 ID to enable

the communication between the two interested peer ARs.

In more detail, all the steps of the inter-domain CARD

process are reported below (refer to Fig. 5). We assume that

ARh belongs to NSP1 and ARz belongs to NSP2.

1. When an MN managed by APh,s enters the coverage area

of APz,k, if the involved APs use the same RAT, the MN

listens to the beacons transmitted by APz,k.

2. The MN notifies ARh of the L2 ID of APz,k.

3. If ARh is not able to resolve the mapping within NSP1, it

sends a unicast request to its CGW (namely, CGW1) with

the L2 ID to be resolved. This message also contains the

invitation for CGW1 to join MGh, the mapping between

the IP address of ARh and the L2 ID of APh,s, and the

relevant service capabilities.

NSP2NSP2

NSP3NSP3

INTERNETINTERNET

Multicast

Unicast

ARh

ARi

ARj

ARk

ARl

ARzCGW1CGW3

CGW2

NSP1SC update

Fig. 6. Inter-domain CARD steady phase.

D. Di Sorte et al. / Computer Communications 29 (2006) 337–357 345

4. CGW1 forwards (in unicast) the query towards the

CGWs of those NSPs that may have a radio coverage

overlapped by that of NSP1.

5. CGW2 finds out that the L2 ID belongs to an AP

managed by ARz, which is an AR of NSP2. At this stage,

if a local multicast group associated with ARh (the one

that issued the request to CGW1) does not already exist,

CGW2 will create it in NSP2, named MGARh/NSP2; such

a group will be used for future service capability updates

from ARh within NSP2. Then, CGW2 forwards the query

to ARz, inviting it to join to MGARh/NSP2, and

communicating the service capabilities associated with

the (ARh,APh,s) pair.

6. ARz replies to CGW2 with a message containing the

mapping between the IP address and the L2 ID, the

relevant service capabilities, and the invitation for CGW2

to join its local multicast group MGz. In addition, ARz

joins MGARh/NSP2.

7. CGW2 forwards the info from ARz to CGW1 and joins

MGz.

8. If a local multicast group associated with ARz (named

MGARz/NSP1) does not already exist, CGW1 will create

one in NSP1, and then forward the message with ARz

information to ARh, inviting it to join this (new) local

multicast group (MGARz/NSP1). Then, CGW1 joins

MGh.

9. Finally, ARh joins this new local multicast group

(MGARz/NSP1).

As regards the steady phase, this is executed exploiting

inter-domain, unicast signalling among CGWs and the new

multicast groups within NSPs (MGAR/NSP).

We assume that ARi, ARj, ARk, ARl and ARz have

wireless coverage overlapping with ARh, and thus have to

receive its updates (refer to Fig. 6). ARi, ARj, and ARh

belong to NSP1, whereas ARk and ARl belong to NSP3 and

ARz belongs to NSP2. After the completion of the discovery

phase, ARi, ARj and CGW1 belong to MGh, whereas ARz

belongs to MGARh/NSP2, and ARk and ARl belong to

MGARh/NSP3. Let us focus on a CGW. If an incoming

message is an SC update through an intra-domain multicast

message from an AR, the involved CGW has to forward it as

a unicast message to the appropriate CGWs. On the other

side, if the incoming message is an inter-domain unicast

update from another CGW, the relevant information is

multicasted within the NSP through the appropriate multi-

cast group. In Fig. 6, solid lines represent multicast

transmissions, whereas dashed lines are associated with

unicast messages.

Clearly, SAs among CGWs must be implemented to

ensure authentication, integrity and confidentiality of data

exchange.

We underline that the depicted procedure might also be

useful to manage a single, large administrative domain as a

number of smaller, co-ordinated sub-domains.

4. The TAR approach

As regards the TAR selection procedure, we assume that

it is performed at the current AR. This implies (i) a

substantial power saving at MNs, (ii) highly complex

terminal equipment is avoided, (iii) critical service

information is managed only among ARs (security issues),

(iv) bandwidth waste is avoided on wireless links, (v) the

possibility to dynamically set specific traffic management

policies for the operator.

D. Di Sorte et al. / Computer Communications 29 (2006) 337–357346

4.1. Handover management

TAR selection may be triggered by three different events:

1. at an explicit request by the customer;

2. periodically, as decided by the current AR, with period

TTAR, which can be dynamically adjusted by the AR on

the basis of the current load, with the aim of spreading it

evenly over the APs. A viable proposal is that each MN is

assigned a period equal to

TTAR Z max TD 1 KeKdSCC

� �; TTAR;min

n o(1)

where SC is the service capability, i.e. the amount of

bandwidth available in the AP, C is the capacity of the

AP, TD and d are design parameters. The higher the

available network resources, the longer the time period

TTAR. The value of TTAR is first determined at connection

set-up, then it is updated at each TAR event according to

the current load of the AR–AP pair the MN is attached to.

When the amount of traffic increases, the frequency of

TAR events also increases (up to a maximum fixed value

of 1/TTAR,min) with the aim of increasing the reactivity of

the CARD/TAR procedure to track system changes.

Clearly, this implies a more aggressive load balancing

action, at the cost, however, of increased complexity in

terms of signalling and computational burden;

3. at an explicit request from the MN, when it detects that

the power level of the received signal is rapidly

decreasing. In particular, this request is sent when the

received power PW is below a threshold PWopt. We

consider the MN to be outside the coverage area when

PW falls below the receiver sensitivity, PWmin.

Upon a TAR event, the MN is required to send its

handover preferences to the current AR. In our scenario,

three types of information are sent as preferences:

1. which types of RAT the MN is able to hand over to (this

information could also be maintained by the AR in the

MN context);

2. the set of the L2 IDs of the APs (of the same RAT of the

one currently active in the MN) the beacons of which

have been received by the MN during the final Tbeacon

seconds. We refer to this set of information as the TAR

beacon list;

3. the received power level PW of the surrounding APs of

the TAR beacon list.

If different ARs are CARs, a subset of them will have

to be selected by the current AR as TARs. The proposed

TAR algorithm aims to select a subset of maximum N

CARs (N is a design parameter), by means of a metric.

The current AR notifies the MN of the list of wireless

accesses to hand over to. At this point, the MN can try

the handover.

We remark that the TARs are not only the APs whose

beacons have been communicated by the MN (i.e. those

with the same RAT as the one currently active in the MN),

but also the APs with other RATs, the coverage areas of

which are supposed to overlap with that of the current AP.

As mentioned above, this information is inferred from the

relevant values of probability of achieving a successful

handover stored in the CARD table of the current AR.

If the MN performs a layer 2 inter-technology handover,

the involved AR(s) must be made aware of any successful

and failed inter-technology handover to update the value of

the handover success probability in the CARD table. In the

case of a layer 3 handover, the new AR sends a positive

acknowledgement to the old AR whenever a handover of a

previously handled MN is successful. In the case of intra-

AR layer 2 handover, the current AR does not change, and it

recognises the layer 2 handover between two of its APs. As

regards any failed handover attempt, the old AR does not

receive any positive acknowledgement, thus it is also able to

infer these failures from the TAR list communicated to the

MN and from the positive acknowledgement it may have

received. It is clear that the additional processing and

signalling burden is the price to pay in order to make the

procedure capable of managing not only intra-technology,

but also inter-technology handovers.

In the following sub-section, we describe the TAR

selection algorithm, the aim of which is to provide the N

most appropriate CARs for the handover.

4.2. The TAR selection algorithm

The metric used represents the criterion which provides

the quality measure MTAR(APh,s,APz,k), of a supposed

handover from the current APh,s to a target APz,k. TAR

driven handovers may occur only towards the APs which

have an entry in the tables managed by the current AR. Our

proposal is that the score associated with a candidate ARz–

APz,k pair for a handover from the APh,s is given by

MTARðAPh;s;APz;kÞ Z f1ðSCz;kÞf2ðPWz;kÞf3ðPrAPðh;sÞ/APðz;kÞÞ

(2)

where the proposed functions fi, iZ1,2,3 are:

f1ðSCz;kÞ Z min 1; eb

SCz;kCðK1ÞjB

C

� �K1

( )if SCz;k R jB

0 otherwise

8><>:

(3)

f2ðPWz;kÞ Z 1 KeKg

PWz;kKPWminPTKPWz;k

� �if PWz;k OPWmin

0 otherwise

8<:

(4)

f3ðPrAPðh;sÞ/APðz;kÞÞ Z PrAPðh;sÞ/APðz;kÞ ZNHO; SUCC

NHO; TOT

(5)

D. Di Sorte et al. / Computer Communications 29 (2006) 337–357 347

Since all the three proposed functions range between 0

and 1, the final score will also range between 0 and 1.

With reference to Eq. (3), B represents the MN

bandwidth demand to satisfactorily support the current

communication session, SCz,k is the service capability of

APz,k, C is the capacity of APz,k, and b is a design parameter.

In order to avoid ping-pong effects, the value of j is set to 1

for all but the current candidate, for which jZ0. The higher

the value of b, the higher the score associated with the

bandwidth (SCC(K1)jB) available after the hypothetical

execution of the handover, normalised by the capacity of the

AP taken into account. Moreover, if the new network access

cannot accommodate such a traffic flow with the necessary

bandwidth, its score is zero. This is a sort of admission

control function executed at the AR.

With reference to (4), PT is the standard value of the

transmission power associated with a given RAT, PWz,k is

the received power from the APz,k, and g is a design

parameter similar to b. Clearly, the higher the value of g, the

higher the sensitivity of the score function to power levels

below the optimum value PWopt.

With reference to (5), PrAP(h,s)/AP(z,k) is the estimation

of the handover success probability from APh,s to APz,k.

This probability value should approximate the percentage of

the APh,s coverage area which overlaps the APz,k coverage

area, thus a low score is given to the candidate APs with low

coverage areas overlapping with that of the current AP. This

parameter helps us in managing inter-technology handovers

without relying on coverage information from MNs. It is

worth noting that, for APs of the RAT currently used, f3 is

always set to 1 (since the overall process relies on beacon

detection), whereas, in the case of different RATs, f2 is

always set to 1 (since the MN does not estimate any power

level).

The set of handover management events includes all the

TAR executions (which may or not result in a handover

attempt) and non-assisted handovers. We define as a failure

event a handover management action that would not be

needed if the CARD/TAR functions worked properly. Thus,

at a given time, the probability, PrF, that a handover

management event fails is equal to

PrF Z PrF1 CPrF2 (6)

where PrF1 is the probability of having a non-assisted

handover (i.e. the TAR process has not been able to provide

a timely, appropriate solution), and PrF2 is the probability

that the MN does not succeed in attaching to the TARs (i.e.

the selection process forces the MN to try to connect to other

wireless accesses which are out of the coverage of the MN).

In other words, there are a couple of reasons for a non-zero

value of the failure probability of the handover management

process. Firstly, a failure may be caused by the movement of

MNs. In fact, upon a notable change of position, the wireless

accesses suggested by the TAR may be aged, so that the MN

may remain out of coverage. Then, another component of

the failure probability is due to the intrinsic characteristic of

the procedure, which, for inter-technology handovers, is

blind to the movement direction of MNs which are driven in

handing over.

In the following we will refer to PrF as the failure

probability of the CARD/TAR procedure.

Let us now illustrate the algorithm to select the list of

N TARs. In order for the selection procedure to make

sense, a CAR is selected only if its score is higher than

that of the current AR. In principle, the N candidates

with the highest score should be picked up. However, if

there are CARs of the current RAT with a score higher

than the current AR, the best of them is inserted at least

at the Nth rank of the handover list, independently of its

score. This policy enables to lower the probability of

handover failure after scanning all the N TARs provided

by the current AR. In fact, intra-technology attempts rely

on the results of previous L2 ID scanning process by the

MN.

5. Numerical results

In this section, we present some of the results obtained

from a simulation trial, the aim of which is (i) to test the

effectiveness of the proposed CARD/TAR approach in

terms of handoff success probability, and (ii) to test the

sensitivity of performance to variations of system

parameters.

We start by illustrating the main features of the

simulation model adopted. Then, we give details about the

simulation scenario, and finally we discuss the numerical

results.

5.1. Simulation model

In this sub-section, we describe the design approach and

the basic implementation choices to develop the simulator

used to evaluate the performance of the CARD/TAR

mechanism.

The simulator is flow-oriented and event-driven, where

packet transmissions are not simulated. In other words,

communications of signalling information occur instan-

taneously, and no delays are associated with information

transfer between network entities. The amount of access

network capacity is managed as a logical resource, part of

which may be allocated to an MN upon successful traffic

flow set-up.

We developed an object-oriented simulator by using the

CCC programming language. In more detail, we used the

event scheduler of the NS-2 simulator [24], and we added

our own CCC modules to manage MN mobility, CARD

functions, and TAR selection. In addition, we used the

OTcl interface of NS-2 to configure the simulation with our

own Tcl scripts. The set-up of the simulation was managed

by means of a Tcl script (typical of the NS-2 environment),

NS-2

SCHEDULERTclCL

ProprietaryTcl libraries

ProprietaryC++ libraries

INPUT FILES MATLABPOST-PROCESSING

SIMULATIONSCRIPTS

OUTPUT FILES

Fig. 7. Simulator architecture.

D. Di Sorte et al. / Computer Communications 29 (2006) 337–357348

which retrieves all the simulation parameters (relevant to

the network topology, RAT features, mobility model,

traffic load and TAR metric) by a number of input files

with a fixed structure and passes them to the CCCmodules by means of the CCC/OTcl interface. Output

files are then processed with Matlab to obtain performance

curves. The architecture of the simulator is depicted in

Fig. 7.

The events which populate the scheduler and imply the

update of the state variables are (i) the updates of the

positions of MNs, (ii) the set-up and tear-down of traffic

flows, and (iii) periodic events associated with the

CARD/TAR procedure (e.g. periodic CARD table

updates).

The coverage areas of APs are assumed to be circular

with a maximum radius denoted as rmax. The power

threshold PWopt corresponds to the one perceived at a

distance equal to 75% of the maximum coverage radius, and

it is calculated by using a quadratic attenuation law. This

simple propagation model does not invalidate the generality

of the CARD/TAR results. In fact, in order to analyse the

proposed algorithms, it is important to define the boundaries

of the coverage areas and the power threshold, regardless of

the specific propagation laws, the study of which is beyond

the scope of this paper. We implemented the coverage of an

AP as hard: so long as the received power level is higher

than PWmin, the MN is in coverage, whereas it is

disconnected when the received power level drops just

below that value.

The capacity, C, associated with an AP is considered as

net capacity, i.e. we assume ‘ideal’ layer 2 technologies,

without considering either the details relevant to trans-

mission aspects (e.g. fading and shadowing), or MAC

protocol features.

As regards the mobility model, we implemented the

Gauss–Markov model [12]. We used this model since it

avoids sharp direction changes, by allowing previous speed

and direction to influence future mobility. The equations

controlling MNs mobility are:

sn Z asnK1 C ð1 KaÞsave Cffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffið1 Ka2Þ

psxnK1

dn Z adnK1 C ð1 KaÞdave Cffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffið1 Ka2Þ

pdxnK1

8<: (7)

xn Z xnK1 CsnK1 cosðdnK1Þ

yn Z ynK1 CsnK1 sinðdnK1Þ

((8)

where sn and dn are the speed and the direction of the MN

at time n, 0%a%1 is the tuning parameter used to modify

the random nature of the motion, save and dave are constant

values representing the average values of sn and dn, sxnK1

and dxnK1are Gaussian random variables with zero mean

and variance l and, b, respectively. Clearly, xn and yn

represent the surface co-ordinates of the MN position in

the area at time n. The position of an MN is updated with

the step fixed at 1 s. The default value of the average

speed was set equal to 1.5 m/s, whereas the parameter a

was set to 0.5 and b was set to p/2. The initial value of

the direction was set to 0. However, when a MN reaches a

distance from the border of the simulation area inferior

than the 5% of the area side, this value is shifted by p (or

by Gp/4 if the MN reaches one of the corners of the

simulation area).

We assume that an MN is instantaneously aware it is

under the coverage of an AP (i.e. it immediately listens to

the relevant beacons).

The network is loaded with homogeneous CBR flows

with bandwidth equal to B. Calls are generated with a

Poisson arrival process with an average frequency l, and the

call duration is modelled as an exponentially distributed

random variable with mean 1/m. An MN can be in one of the

three following states: (i) disconnected (DISC), if it is not

under the coverage of any AP; (ii) connected (CONN), if it

is under the coverage of an AP; (iii) in communication

(COMM), if it is under the coverage of an AP and has an

active communication session, and therefore uses an amount

of the AP bandwidth equal to B.

Table 2

Characteristic parameters of radio access technologies

RAT parameters RAT1 RAT2 RAT3

f (frequency) 5.4 GHz 2.4 GHz 2 GHz

rmax (maximum radius) 20 m 30 m 40 m

C (capacity) 10 Mb/s 5 Mb/s 2 Mb/s

GT (transmission gain) 1 1 1

GR (reception gain) 1 1 1

PT (transmitted power) 100 mW 100 mW 100 mW

PWopt (power threshold) 8.7 nW 19.5 nW 15.8 nW

PWmin (receiver sensitivity) 4.9 nW 11 nW 8.9 nW

D. Di Sorte et al. / Computer Communications 29 (2006) 337–357 349

Before running our simulations, we validated all the

features added to the base version of NS-2. In more details,

we executed the following steps:

† validation of the Gauss–Markov mobility model by

comparing the characteristics of MN trajectories

obtained by our implementation with those presented

in [12];

† validation of the dynamics of the CARD table

construction by analysing the discovery events triggered

by a number of MNs with deterministic trajectories;

† validation of the CARD table data in the steady phase by

an exhaustive comparison with the overlapping of the

radio coverage provided by the APs;

† validation of the traffic generation process and band-

width allocation mechanism in APs by analysing the

output results of simulations with a number of MNs with

deterministic trajectories;

† validation of the TAR mechanism by means of (i) a

check of the (AR,AP) pair an MN is attached to before

and after a TAR event, and (ii) a consistency check

between the TAR scores associated with the (AR,AP)

pairs around a given MN, as well as the actual position of

the MN and resource availability of the accesses.

The performance curves shown in Section 5.3 are

obtained by averaging 60 simulation runs. The (small)

confidence intervals are not shown to improve the neatness

of the figures.

5.2. Simulation scenario

The network topology simulated consists of a single

administrative domain with a number of APs and ARs. We

envisage the presence of three different RATs. The main

peculiarities of RATs are reported in Table 2.

The whole network area is a square with a side measuring

150 m. We ran simulations over three different coverage

maps.

In the first coverage configuration (referred to as ‘sparse’

coverage), all the area is covered by at least one RAT. This

assumption implies that only a multi-mode terminal can

move anywhere within the area without losing the wireless

connectivity. The number of APs is 44 (27 of RAT1, 11 of

RAT2, 6 of RAT3) and the number of ARs is 8.

The second coverage (referred to as ‘medium’ coverage)

is characterised by a higher amount of wireless access

resources than in the previous one. The total number of APs

is 50 (30 of RAT1, 13 of RAT2, 7 of RAT3) and the number

of ARs is 10.

Finally, the third network configuration (referred to as

‘dense’ coverage) is characterised by a coverage which

ensures that the whole area is covered by at least one AP for

each RAT; this means that each RAT provides a full coverage

of the area. Consequently, also a single mode terminal can

move anywhere within the area without losing the wireless

connection. The number of APs is 72 (42 of RAT1, 19 of

RAT2, 11 of RAT3) and the number of ARs is 14.

The association AR–APs in the network was designed

according to the following criteria: (i) an AR has to be

connected to APs of different RATs (at least 2); (ii) APs

connected to the same AR should be very close to each other,

possibly with overlapping coverage; (iii) the minimum

distance between two APs of the same RAT is equal to the

maximum radius reported in Table 2. This condition is used

to limit the coverage overlapping of APs with the same

technology. In addition, the coverage maps were designed so

that the wireless network resources are uniformly distributed

over the area. The output capacity of ARs was set at 45 Mbps.

Simulation results are relevant to homogeneous flows.

Each flow is associated with a bandwidth value, B, equal to

128 Kbps. The number of MNs is 1800. In order to test the

procedure and to limit the number of rejected calls and failed

handovers due to lack of resources (especially in the middle

of the square area where the density of MNs is higher than at

the borders), we loaded the network with an average amount

of traffic equal to 30% of the total capacity. We set the

average call duration equal to 5 min (i.e. mZ1/300 sK1), thus

l is equal to 2.50, 2.96, and 4.2 sK1, corresponding to an

average load of 740, 888, and 1258 Erlangs, relevant to the

sparse, medium, and dense coverage, respectively. As

regards the TAR selection process, we ran three simulations

with a maximum number (N) of handover attempts towards

TARs equal to 2, 3 (default value), and 4, respectively.

The TAR selection is triggered by either the MN, when

the received power level from the current AP drops below

the threshold PWopt, or periodically by the current AR, with

period TTAR, computed from (1), with values of TD equal to

30 (default value), 60, and 90 s, TTAR,minZ0.1TD seconds,

and dZ2 ln 10 (so that TTAR(SC/CZ0.5)Z0.9TD).

We also analysed the sensitivity of the model with respect

to MN speed, set equal to 0.5, 1 and 1.5 m/s (default value).

The power control in MNs is assumed to be executed

every second.

The parameters b and g in the TAR function are selected

according to the following criteria:

† for each RATi, iZ1,2,3, the value of b is set in such a

way that, when the amount of the available bandwidth

(SC-B, for candidate ARs, and SCCB, for the current

100

101

0

0.2

0.4

0.6

0.8

1

Service Capabilities (Mb/s)

f 1(SC

)

RAT1, C = 10 Mb/s

RAT2, C = 5 Mb/s

RAT3, C = 2 Mb/s

5 10 15 20 25 30

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

Received Power Level (PW) in nW

f 2(P

W)

RAT1, γ = 36 106, PW

opt = 8.7 nW

RAT2, γ = 16 106, PW

opt = 20 nW

RAT3, γ = 20 106, PW

opt = 16 nW

Fig. 8. Metric functions f1 and f2.

D. Di Sorte et al. / Computer Communications 29 (2006) 337–357350

one) is higher than, or equal to 0.8Ci, Ci being the full

capacity of RATi, then the weight associated with the

service capability is equal to 1. This implies that the

value of b is equal to bZ0.866 for all RATs. For values

of SC lower than 0.8Ci, such a weight rapidly decreases,

and consequently the importance of the load balancing

criterion increases;

† for each RATi with iZ1,2,3, the value gi is set so that the

function f2(PWopt,i)Z3/4; in other words, when the

power level is equal to the value of PWopt,i which

0 500 1000 150.9

0.95

1

Simulation tim

Han

dove

r su

cces

s pr

obab

ility

N = 3T

D = 30 s

speed = 1.5 m/

Fig. 9. Probabilities of successful ha

characterises the RATi itself, we set the weight

associated with the received power level equal to 3/4.

For values of received power lower than PWopt,i, this

weight rapidly decreases. Consequently, the power-

based criterion becomes important when the received

power level from an AP is lower than PWopt. The final

result gave values of g as g1Z36.5!106, g2Z16.2!106, and g3Z20!106, for RAT1, RAT2 and RAT3,

respectively. In this way, we take into account different

characteristics of RATs, i.e. the signal strength that, as

00 2000 2500 3000e (seconds)

P1, Dense cov

P2, Dense cov

P3, Dense cov

s

ndover: dense coverage case.

D. Di Sorte et al. / Computer Communications 29 (2006) 337–357 351

all channel specific factors, cannot be straightforwardly

compared.

The curves relevant to f1 and f2 are reported in Fig. 8.

As regards the update of the service capabilities, ARs

multicast the update of the available bandwidth, SC, to an

AP either periodically, with period equal to TSCZ60 s, and

upon variations of SC of at least 5% from the one previously

communicated.

5.3. Simulation results

Our goal is to verify the effectiveness of the

CARD/TAR procedure, in particular as regards the

capability of not only self-constructing the coverage map

of the whole area, but also of driving both inter and intra-

technology handovers.

Initially, the CARD table stored in a generic AR contains

only the information relevant to the APs connected to the

AR itself, then, as time goes by, this table self-constructs

according to the information provided by the MNs. As

regards the data collected to generate statistics, we consider

only those relevant to the MNs in the COMM state.

Let us first analyse the simulations in the case of dense

coverage, with TDZ30 s, MN speed equal to 1.5 m/s, and

NZ3.

0 500 1000 150

0.2

0.4

0.6

Simulation tim

Fai

lure

pro

babi

lity

N = 3T

D = 30 s

speed = 1.5 m/s

Fig. 10. Failure probabilities

Fig. 9 shows the cumulative probability of successfully

handing over within the first, second and third attempt. As

expected, we can see that after a brief transient period

needed for self-constructing the CARD tables, the prob-

ability of successfully handing over towards the best-scored

CARs rapidly converges to values close to 1. Clearly, the

successful handover probability increases with the number

of attempts.

As regards the efficiency of the procedure to drive

handovers, the failure probability of the CARD/TAR

procedure versus simulation time is depicted in Fig. 10.

We compute the failure probability of the overall

process to manage handovers, PrF, as the number of failure

events (defined in Section 4.2) over the number of

handover management events (i.e. the number of TAR

process executions plus the number of non-assisted

handovers).

In the case of dense coverage, PrF is approximately 5.6%.

This value is the result of two different contributions (see

(6)). The former, PrF1, is equal to 5.27%, whereas the latter

is equal to 0.33%. Note that, considering a single run, out of

a total number of 186165 successful handovers, 149114 of

them (80%) are driven by the procedure. It is worth noting

that a high number of TAR events provide an empty TAR

list so that the MN is advised to remain under the current

coverage (about the 77% of TAR executions).

00 2000 2500 3000e (seconds)

Inter-technology, dense covGlobal, dense cov

: dense coverage case.

D. Di Sorte et al. / Computer Communications 29 (2006) 337–357352

Fig. 10 also shows that the probability of failing an inter-

technology handover is approximately 20% at the end of the

simulation, when the probability values in the CARD tables

become stable.

We note that the performance of the CARD/TAR

procedure is extremely satisfactory for dense wireless

coverage. This result is to be expected for two main reasons:

1. inter-technology handovers (the most critical ones) are

not so frequent since:

a. each RAT covers the whole area;

b. the discovery of the surrounding coverage provided

by the current RAT depends on the MN capability to

pick up the L2 beacons transmitted by the

surrounding APs. In general, we can expect the

mechanism to work better when the overlapping

area between neighbour APs belonging to the same

RAT is substantial;

c. the traffic load on offer is not heavy. This also means

that inter-technology handovers are not so often

solicited by a load-balancing action.The distribution

of successfully driven handovers, classified as intra

and inter-technology handovers, together with their

distribution over the three different RATs, is

summarised in Table 3 for a single run. Driven

intra-technology handovers are equal to 92.28% of

the total driven handovers and equal to 73.91% of

the total handovers. As expected, most handovers

occur within RAT1, which has the largest capacity.

On the other hand, driven inter-technology hand-

overs are equal to 7.72% of the total driven

handovers and equal to 6.18% of the total handovers;

2. overlapping between the coverage areas of different

RATs is high enough to keep the failure probability of

inter-technology handover attempts sufficiently low as

not to degrade the performance of the overall procedure.

Consequently, if we consider a coverage with an amount

of resources lower than those in the dense coverage, a

deterioration of the performance of the procedure is to be

expected.

Let us analyse what happens in the case of medium and

sparse coverage. Figs. 11–13 show the cumulative

Table 3

Driven Handover distribution: dense coverage case

Handover Type Involved RATS % Total %

Intra-technology RAT1/RAT1 66.43 92.28

RAT2/RAT2 20.53

RAT3/RAT3 5.32

Inter-technology RAT1/RAT2 2.31 7.72

RAT1/RAT3 0.27

RAT2/RAT1 3.1

RAT2/RAT3 0.87

RAT3/RAT1 0.24

RAT3/RAT2 0.93

probability of a successful handover within the first, second

and third attempt, respectively, in all the coverage cases. It

is clear that performance improves from sparse to medium

coverage and from medium to dense coverage. This trend is

highlighted in Fig. 14, where the probability of failure

increases when compared with the dense coverage case.

This is due to a lower amount of wireless network resources,

hence to a reduced overlapping of coverage areas of

different RATs. As a result, the percentage of driven

handovers remains nearly constant for medium coverage

(99995 out of a total of 124681, i.e. 80%) and rapidly

decreases for sparse coverage (67695 out of a total of

93005, i.e. 72.8%). Note that the number of handovers for

MNs in the COMM state is higher in the case of dense

coverage since the amount of active flows is higher than in

the other two cases. As regards the percentage of inter-

technology driven handovers, it generally rises when the

coverage area provided by a single RAT decreases; it is

equal to 11.54% in the case of sparse coverage. In addition,

the probability of failing the CARD/TAR process increases

from 5.6% for dense coverage up to approximately 6% for

medium and 7.2% for sparse coverage (PrF1 is equal to

5.5%, whereas PrF2 is equal to 1.7%). We remark that, as

expected, PrF2 has increased considerably with respect to

the dense coverage case, since inter-technology handovers

are needed more frequently.

Let us now analyse the performance of the procedure in

the case of sparse coverage when N ranges in the set {2, 3,

4}, TD in the set {30, 60, 90 s}, and the average speed of the

MNs in the set {0.5, 1, 1.5 m/s}. We expect a performance

improvement when N increases, TD decreases, and MN

speed decreases. In fact, lower values of TD imply more

frequent TAR actions, thus a better performance of the

CARD/TAR procedure, which is able to quickly track the

network changes and the MN position. Moreover, by

increasing the value of N means that the TAR list, provided

to the MNs by the current AR, includes a higher number of

CARs to try to hand over to, and this implies an

improvement in the performance of the procedure.

Furthermore, we expect the process to work better when

the speed of MNs is decreased, since the probability of

having a non-assisted handover due to large movement will

thus decrease. Note that the values of the MN speed

(walking speed) that we have chosen are typical of the

network scenario simulated, where the radius of the APs

ranges from 20 to 40 m.

Fig. 15 shows the cumulative probability of a successful

handover within the first N handover attempts for cases NZ2, NZ3, and NZ4, with TDZ30 s and MN speed equal to

1.5 m/s. The improvement is clear for each handover

attempt. We note in particular the improvement of the

successful handover probability at the last attempt. This

value varies from 0.885 for NZ2, to 0.905 for NZ3, and,

finally, to 0.91 for NZ4. This improvement induces only a

negligible decrease in the failure probability of the

CARD/TAR procedure.

0 500 1000 1500 2000 2500 30000.7

0.75

0.8

0.85

0.9

0.95

1

Simulation time (seconds)

Han

dove

r su

cces

s pr

obab

ility

at t

he fi

rst a

ttem

pt

N = 3T

D = 30 s

speed = 1.5 m/s

P1, Dense cov

P1, Medium cov

P1, Sparse cov

Fig. 11. Probability of successful handover at the first attempt: dense, medium, and sparse coverage.

0 500 1000 1500 2000 2500 30000.82

0.84

0.86

0.88

0.9

0.92

0.94

0.96

0.98

1

Simulation time (seconds)

Han

dove

r su

cces

s pr

obab

ility

at t

he s

econ

d at

tem

pt

P2, Dense cov

P2, Medium cov

P2, Sparse cov

N = 3T

D = 30 s

speed = 1.5 m/s

Fig. 12. Probability of successful handover within the second attempt: dense, medium, and sparse coverage.

D. Di Sorte et al. / Computer Communications 29 (2006) 337–357 353

0 500 1000 1500 2000 2500 30000.8

0.82

0.84

0.86

0.88

0.9

0.92

0.94

0.96

0.98

1

Simulation time (seconds)

Han

dove

r su

cces

s pr

obab

ility

at t

he th

ird a

ttem

pt

P3, Dense cov

P3, Medium cov

P3, Sparse cov

N = 3T

D = 30 s

speed = 1.5 m/s

Fig. 13. Probability of successful handover within the third attempt: dense, medium, and sparse coverage.

0 500 1000 1500 2000 2500 30000

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

Simulation time (seconds)

Fai

lure

pro

babi

lity

Dense cov, inter-technologyDense cov, globalMedium cov, inter-technologyMedium cov, globalSparse cov, inter-technologySparse cov, global

N = 3T

D = 30 s

speed = 1.5 m/s

Fig. 14. Failure probabilities: sparse, medium, and dense coverage.

D. Di Sorte et al. / Computer Communications 29 (2006) 337–357354

1 1.5 2 2.5 3 3.5 40.8

0.82

0.84

0.86

0.88

0.9

0.92

0.94

Handover attempt index

Han

dove

r su

cces

s pr

obab

ility

Sparse cov, N=2Sparse cov, N=3Sparse cov, N=4

TD

= 30 s

speed = 1.5 m/s

Fig. 15. Probability of successfully handing over versus handover attempts for NZ2,3,4: sparse coverage.

30 40 50 60 70 80 900.82

0.83

0.84

0.85

0.86

0.87

0.88

0.89

0. 9

0.91

TD

(seconds)

Han

dove

r su

cces

s pr

obab

ility

Sparse cov, P1

Sparse cov, P2

Sparse cov, P3

N = 3speed = 1.5 m/s

Fig. 16. Probability of successfully handing over within the first, second, and third attempt versus TD: sparse coverage.

D. Di Sorte et al. / Computer Communications 29 (2006) 337–357 355

0.5 0.6 0.7 0.8 0.9 1 1.1 1.2 1.3 1.4 1.50.01

0.02

0.03

0.04

0.05

0.06

0.07

0.08

MN speed (m/s)

Failu

re p

roba

bilit

ySparse cov, globalSparse cov, Pr

F1Sparse cov, PrF2

N = 3T

D = 30 s

Fig. 17. Failure probabilities of the CARD/TAR process versus MN speed: sparse coverage.

D. Di Sorte et al. / Computer Communications 29 (2006) 337–357356

In addition, as expected from the above considerations,

the cumulative probability of successfully handing over

within the first, second and third attempt (N is fixed to 3 and

MN speed is equal to 1.5 m/s) slightly improves when TD

decreases, as shown in Fig. 16. The relevant improvement in

terms of failure probability is negligible.

Finally, Fig. 17 shows the strong improvement of the

failure probabilities (PrF, PrF1, and PrF2) of the CARD/TAR

process when the MN speed decreases (N is fixed to 3 and

TD to 30 s). It is worth noting that, as expected, the

improvement of the performance of the process is mainly

due to the fact that PrF1 lowers when the average MN speed

decreases. In fact, in this case, the number of non-assisted

handovers lowers considerably, since the CARD/TAR

process is able to better follow the MN movement. Clearly,

similar improvements have been verified also in the dense

and medium coverage scenarios. We stress that the effect of

the MN speed is much stronger than the one of N and TD.

We can conclude that the proposed approach works very

well even in the sparse coverage scenario, under the

assumption that MN speed is not too high.

6. Conclusions

We considered a framework in which an administrative

domain provides wireless access by means of heterogeneous

radio technologies. In this scenario, we proposed an intra-

domain distributed CARD/TAR procedure based on L2

beacon listening, to (i) dynamically self-construct a local

map at each AR of the surrounding wireless coverage,

together with the relevant capabilities, and (ii) select the

most suitable access to the network.

A novelty of the PMM CARD approach is the use of

push-mode multicast transmissions. This enables the

discovery phase to be speeded up and the signalling burden

to be lowered in the steady phase; a quantitative analysis of

these performance figures is presented in [22].

Another important aspect is that the proposed CARD

procedure is also able to provide the TAR algorithm with

the necessary, updated information to drive network-

assisted handovers at layers 2 and 3, both intra and inter-

technology. This algorithm, used to select the best AR–

AP for handing over to, is based on a set of criteria

which includes bandwidth availability and received

power level. The procedure is generally valid for multi-

mode terminals. In particular, we considered the very

challenging situation of multi-mode terminals with a

single network interface turned on. To infer knowledge

about the wireless coverage offered by the RATs

different from the one currently active on MNs, we

introduced a statistical parameter within the TAR metric

which is the probability to execute a successful inter-

technology handover.

D. Di Sorte et al. / Computer Communications 29 (2006) 337–357 357

We analysed the performance in terms of handover

success and failure probability of the CARD/TAR pro-

cedure by simulations. In particular, we have verified that

the procedure is very effective when the size of the

overlapping coverage areas among APs is considerable. In

this case, most of handovers are driven by our CARD/TAR

procedure, and inter-technology handovers, which are

clearly the critical ones, are not that frequent. The

probability of failure of inter-technology handover attempts

is sufficiently low and it does not significantly affect the

failure probability of the CARD/TAR process. In addition,

when the average speed of MNs is limited, we have shown

that the proposed approach works very well even in a sparse

coverage scenario.

Note that inter-technology handovers are generally

highly power consuming and potentially critical when

real-time services are supported. We stress that this

criticality is mainly due to the fact that the selection

procedure is blind and failures may occur. It is clear that, if

multi-mode terminals are allowed to simultaneously scan

for beacons on all network interfaces, the procedure is no

longer “blind”, and failures would be avoided. Note that the

intrinsic CARD/TAR characteristics (quick self-construc-

tion of coverage mapping, effective capability exchange,

load balancing) would remain unaltered. The price to pay

would be an increased complexity and power consumption

in terminals.

Acknowledgements

The authors would like to thank the anonymous referees

for their suggestions and comments that have improved the

quality of the paper. This work was performed under co-

financing by the Italian Ministry of Education, University,

and Research (MIUR) within the FIRB project PRIMO.

References

[1] J. Manner, M. Kojo, Mobility related terminology, IETF RFC 3753,

June, 2004.

[2] M. Liebsch, Candidate access router discovery, Internet draft,

September, 2004.

[3] J. Loughney, M. Nakhjiri, C. Perkins, R. Koodli, Context transfer

protocol, Internet draft, August, 2004.

[4] J. Kempf, Problem description: reasons for performing context

transfers between nodes in an IP access network, IETF RFC 3374,

September, 2002.

[5] C. Perkins, IP mobility support for IPv4, IETF RFC 3344, August,

3344, 2002.

[6] D.B. Johnson, C.E. Perkins, Mobility support in IPv6, MobiCom’96,

Rye, NY, USA, 1996.

[7] A.T. Campbell, J. Gomez, K. Sanghyo, C. Wan, Z.R. Turanyi,

A.G. Valko, Comparison of IP micromobility protocols, IEEE

Wireless Communications, February, 2002.

[8] D. Funato, X. He, C. Williams, A. Takeshita, M.D. Ali, J. Nakfour,

Geographically adjacent access router discovery protocol, Internet

draft, June, 2002.

[9] D. Trossen, G. Krishnamurthi, H. Chaskar, R.C. Chalmers, E. Shim, A

dynamic protocol for candidate access-router discovery, Internet

draft, March, 2003.

[10] R. Koodli, Fast handovers for mobile IPv6, Internet draft, October,

2004.

[11] K. El Malki, et al., Low latency handoffs in mobile IPv4, Internet

draft, June, 2004.

[12] T. Camp, J. Boleng, V. Davies, A survey of mobility models for ad

hoc network research, Wireless Communication and Mobile

Computing (WCMC) 2 (5) (2002) 483–502.

[13] E. Shim, J.P. Redlich, R.D. Gitlin, Secure candidate access router

discovery, IEEE Wireless Communications and Networking Con-

ference (IEEE WCNC), New Orleans, USA, March, 2003.

[14] IETF MSEC WG Homepage, http://www.ietf.org/html.charters/msec-

charter.html.

[15] S. Glass, T. Hiller, S. Jacobs, C. Perkins, Mobile IP authentication,

authorization, and accounting requirements, IETF RFC 2977,

October, 2000.

[16] C.E. Perkins, P.R. Calhoun, AAA registration keys for mobile IPv4,

Internet draft, October, 2003.

[17] G.P. Pollini, Trends in hand-over design, IEEE Communications

Magazine, March, 1996.

[18] P.M.L. Chan, R.E. Sheriff, Y.F. Hu, P. Conforto, C. Tocci, Mobility

management incorporating fuzzy logic for a heterogeneous IP

environment, IEEE Communications Magazine, December, 2001.

[19] K. Pahlavan, et al., Handoff in hybrid mobile data networks, IEEE

Personal Communications 7 (2) (2000).

[20] N.D. Tripathi, Generic adaptive handoff algorithms using fuzzy logic

and neural networks, PhD Thesis, Virginia Polytechnic Institute and

State University, 1997.

[21] W. Zhang, J. Jahnert, K. Dolzer, Design and evaluation of a hand-over

decision strategy for 4th generation mobile networks, IEEE Vehicular

Technology Conference, Jeju, Korea, 2003.

[22] N. Blefari-Melazzi, D. Di Sorte, M. Femminella, L. Piacentini, G.

Reali, Performance evaluation of a multicast-based solution for

wireless resources discovery, IEEE International conference on

Communications, Seoul, Korea, May 2005.

[23] R. Atkinson, S. Kent, Security Architecture for the Internet Protocol,

IETF RFC 2401, Nov, 1998.

[24] NS Web-site: http://www.isi.edu/nsnam/ns/.