AN ARCHITECTURAL FRAMEWORK FOR HETEROGENEOUS

NETWORKING

Glenford Mapp

, David N. Cottingham

, Fatema Shaikh

, Pablo Vidales

Leo Patanapongpibul

, Javier Balioisian

, Jon Crowcroft

School of Computing Science, Middlesex University, Middlesex NW4 4BT, UK

Computer Laboratory, University of Cambridge, Cambridge CB3 0FD, UK

Deutsche Telekom Laboratories, 10587 Berlin, Germany

Samsung Electronics Research Institute, Staines, Middlesex TW18 4QE, UK

Network Management Research Centre, Ericsson, Ireland

Keywords:

Heterogeneous Networking, Vertical Handover, OSI Model, Mobility.

Abstract:

The growth over the last decade in the use of wireless networking devices has been explosive. Soon many

devices will have multiple network interfaces, each with very different characteristics. We believe that a

framework that encapsulates the key challenges of heterogeneous networking is required. Like a map clearly

helps one to plan a journey, a framework is needed to help us move forward in this unexplored area. The

approach taken here is similar to the OSI model in which tightly deﬁned layers are used to specifyfunctionality,

allowing a modular approach to the extension of systems and the interchange of their components, whilst

providing a model that is more oriented to heterogeneity and mobility.

1 INTRODUCTION

Mobile devices with several interfaces such as new

mobile phones and PDAs will soon become common-

place. We deﬁne the networking issues associated

with this type of device as heterogeneous network-

ing and the devices themselves as hetnet devices.

Heterogeneous networking poses many challenges

in several areas. At the lowest levels, many new ac-

cess technologies including 3G, WiMax and UltraW-

ideBand (UWB) will be supported on hetnet devices.

This expansion of the link and MAC layers has been

accompanied by a contraction in the network layer as

core networks use IP packets to facilitate key services

including telephony, data, and multimedia.

One of the major capabilities of heterogeneous net-

working is that of handover. This is necessary be-

cause the networks that are currently available and/or

their points of attachment may be changing as a mo-

bile node changes its location. Another key issue that

needs to be addressed because of heterogenous net-

working is that of Quality of Service (QoS) in pe-

ripheral networks. This is because different wireless

networks have varying QoS so vertical handover not

only affects the point of attachment but also the QoS

of the link as seen by other entities higher up the pro-

tocol stack. This in turn affects the ability of the net-

work and transport services to deliver effective perfor-

mance since these systems must respond to changes in

QoS in the available channels.

At the higher layers how these QoS changes are

dealt with and what can be done by the system to min-

imize their effects on applications must also be exam-

ined. Alternatively, future applications will be able,

with the help of the system, to structure themselves

to make use of QoS changes in wireless networks.

In addition, new kinds of application environments

which can facilitate new features such as personalisa-

tion, personal area networks and location-based ser-

vices will be built.

In order to address the issues above, the authors be-

lieve that there needs to be a framework that encap-

sulates the key challenges of heterogeneous network-

ing for mobile devices. Like a map clearly helps one

to plan a journey, a framework is needed to help us

move forward in this unexplored area. The approach

taken here is similar to the OSI model in which layers

are used to specify functionality, allowing a modular

approach to the extension of systems and the inter-

change of their components such that different imple-

menations can be easily explored. The authors are

implementing a prototype testbed to explore the pro-

posed architecture.

The rest of the paper is structured as follows: Sec-

tion 2 looks at the OSI model which has been used

as a communication architecture. Section 3 describes

Mapp G., N. Cottingham D., Shaikh F., Vidales P., Patanapongpibul L., Balioisian J. and Crowcroft J. (2006).

AN ARCHITECTURAL FRAMEWORK FOR HETEROGENEOUS NETWORKING.

In Proceedings of the International Conference on Wireless Information Networks and Systems , pages 5-12

 SciTePress

the architectural framework. Section 4 describes

the Cambridge Testbed; Section 5 describes work in

progress, while Section 6 describes the issues that are

still to be investigated. The paper concludes with a

summary in Section 7.

2 THE OSI MODEL

The most well-known framework in data communi-

cations is the OSI model (Zimmermann, 1988) de-

veloped by the CCITT. This is regarded as a refer-

ence model, and not an implementation plan, but it

clearly delineates the functions of the layers to pro-

vide a framework for exchanging data between net-

worked applications.

The authors believe that the OSI model cannot

be considered the dominant model for heterogeneous

networking. It is not wrong; it is simply inade-

quate for heterogeneous networking for several rea-

sons. Firstly in the OSI model, the ﬁrst three lay-

ers, the physical, link and network layers are con-

cerned with the movement of packets between net-

works. The higher layers of the OSI model – the

transport, session, presentation and application lay-

ers – are designed to deal with end-to-end issues be-

tween application processes. Such a horizontal sep-

aration between networking and End-to-End Issues

is no longer sustainable in heterogeneous networks.

There are several reasons for this.

In the OSI model, the network is essentially used

to simply forward packets to their relevant destina-

tion. However, in heterogeneous networking, there

are new functions that the network must also support.

One of these is vertical handover. The point here is

that vertical handover requires frequent and intimate

communication between the mobile host and the net-

work which cannot be simply incorporated into the

OSI model. To go further, vertical handover also in-

volves the reconﬁguration of certain parameters in the

network, such as allowing for the reservation of re-

sources to ensure quality of service in the receiving

network. Such a reality is difﬁcult to model in the

OSI context.

The other major observation is that the OSI model

works well when the characteristics of the networks at

the edge are not overly dissimilar from the core net-

work. The early Internet had Ethernet or Token ring

systems which were basically wired networks capa-

ble of several megabits per second between endpoints.

However, looking at network trends since that time,

the core network and end systems have taken different

evolutionary paths. The core network is being made

faster with the use of technologies such as MPLS and

single-mode ﬁbre optics, while the peripheral systems

are rapidly becoming dominated by the emergence of

wireless technologies that have very different charac-

teristics in terms of latency, bandwidth, availability

and error distribution properties.

In the light of these observations, the authors be-

lieve a new framework is required to better reﬂect

how tomorrow’s heterogeneous networks should be

constructed. This new model is is described in the

following sections.

3 THE ARCHITECTURE

3.1 Layering: A Conceptual

Framework

As outlined in the previous section, the OSI model

does not act as a good conceptual framework for to-

day’s networks. A key point is that this does not in-

validate a layered approach, but rather indicates that it

must be updated. Such an approach must specify the

hierarchy of functionalities, but not prescribe detailed

interfaces between them. The ordering of the hierar-

chy is perhaps the most important consideration; this

is evidenced by the difﬁculties that have been experi-

enced in placing, for example, vertical handover func-

tionality above the transport layer.

We propose an architecture that, like the OSI

model, has seven layers, but that uses a novel hier-

archy of functionalities (Figure 1). In our model, ver-

tical handover, with input in turn from policy man-

agement, is placed below the network transport layer.

Similarly, quality of service is given its own speciﬁc

layer to separate it from both the application and the

network transport modules. This contrasts with cur-

rent approaches where QoS appears to be more of an

“add-on” rather than an integral part of the network

stack.

It is also important to realise that the layering

paradigm does not restrict implementors to rigid mod-

ularisation. A framework is a useful concept, rather

than a detailed design speciﬁcation. Hence, whilst

our model details the necessary ordering of function-

ality, it does not discount the possibility of cross-

layer approaches. Indeed, as detailed in Section 4,

policy management and vertical handover functional-

ity might well be integrated into a single component.

Also, higher layers may well need context not only

from the layers directly beneath them in the hierar-

chy, but also those further down. This can be seen to

be the case for the QoS layer, where hints from the

network abstraction layer would be of great utility.

Hence, we emphasise the need for clear concep-

tual separation, whilst leaving open the possibilities

for vertical integration and trans-layer interfacing in

terms of implementation. In the following sections we

Figure 1: The architectural model.

proceed to describe each of these separate conceptual

layers in turn, using a bottom-up approach.

3.2 Hardware Platform Layer

This layer is used to deﬁne the hardware components

and technologies required to support wireless net-

works. This layer deﬁnes several characteristics, for

example, the electromagetic spectrum required for a

given technology, the different modulation techniques

that may be used, as well as the MAC algorithms for

acquiring and reserving channels. It is recognised that

individual systems may be totally incompatible with

each other. Hence, the layer is composed of vertical

sub-layers, with each sublayer representing a particu-

lar network, e.g. 3G, WLAN, WiMax, etc.

3.3 Network Abstraction Layer

The Network Abstraction Layer is used to support dif-

ferent networking technologies using a common in-

terface. Different wireless device drivers will eventu-

ally be written to map onto this layer. The network

abstraction layer has to do with controlling and main-

taining the network on the mobile node. Recently,

the IEEE convened the 802.21

working group to ex-

amine the possibility of standardising the interface to

different wireless MACs. This work is relevant to the

network abstraction layer of our architecture.

3.4 Vertical Handover Layer

Vertical handover is clearly a key component of het-

erogeneous networking (McNair and Zhu, 2004).

There are two distinct approaches to vertical han-

dover. The ﬁrst is the network-controlled approach

in which the network decides when and how the han-

dover will occur. This means that there are mecha-

nisms in the network that maintain all relevant infor-

mation on the mobile host, including its location rela-

http://www.ieee802.org/21/

tive to different networks, their signal strengths at the

mobile node’s location, and its direction and speed.

However, we do not believe this approach is scalable.

As new wireless networks are added, the information

about all the networks to which the host is currently

attached becomes difﬁcult to maintain. Such an ap-

proach also relies on operators sharing detailed infor-

mation about their networks, a concept that is unlikely

in the current cellular telecommunications environ-

ment.

The second approach is termed client-based han-

dover. With this approach, handover is controlled by

the mobile device. There are clear advantages to using

this approach. Firstly the mobile node will keep up

to date information about its network interfaces and

therefore is in a superior position to decide when han-

dover should occur. Secondly, the mobile node will

also be aware of the state of its TCP connections and

other higher level issues, and therefore these factors

can be taken into account when making the decision

to conduct a vertical handover. Finally, by removing

handover functions from the core network, it should

then be more cost effective to build increasingly ro-

bust core systems.

3.5 Policy Management Layer

A policy management system is needed to evaluate all

the circumstances when handover should occur, tak-

ing into account various factors such as changes in

coverage and signal strength, the state of the network,

and the state of any transport connections associated

with the mobile host.

There are two types of policy management which

are being explored: reactive and proactive. A reactive

policy depends entirely on notiﬁcation from the net-

work abstraction layer about the presence or absence

of networks as the hetnet device changes its location.

Almost all the policy management systems so far re-

ported in the literature (e.g. POLIMAND (Aust et al.,

2003)), including the most recent, PROTON (Vidales

et al., 2005a), have been reactive.

Proactive policies attempt to acquire and use infor-

mation about the likely coverage and signal strength

before the mobile actually reaches a given location.

A key parameter sought from proactive policies is the

Time Before Vertical Handover or TBVH. Knowing

the TBVH allows the higher layers of the protocol

stack to make maximum use of channels that may

soon be unavailable.

Proactive policies can be divided into two types: A

proactive-knowledge-based policy mechanism makes

explicit use of knowledge of the mobile node’s loca-

tion and knowledge about the types of coverage and

signal strengths of each network that the hetnet de-

vice is likely to encounter at that location. These sys-

tems require enhanced location detection and moni-

toring routines as well as a mechanism for ensuring

data on what networks are available at a given location

is accurate and easily accessible. Such data must be

available in a format that allow it to be processed by

hosts with limited resources, and transmitted over po-

tentially low bandwidth networks without signiﬁcant

impact. A few systems (Laasonen et al., 2004)(Soh

and Kim, 2004) have utilised data collected from pre-

vious journeys.

Another approach is a proactive-modelling ap-

proach in which a mathematical model is used to de-

termine the TBVH based on simple geometric cal-

culations. Though less accurate than the proactive

knowledge approach, this approach is ﬂexible and can

be used in simulations as well as in real networks.

There is also a growing need to combine proactive

and reactive policies, with a view to developing an ar-

chitecture where it is possible to choose which policy

to use in a given situation. Hence, when there is accu-

rate information then a proactive policy may be used.

However, when there is no coverage data or the data is

unreliable, then the system can fall back to a reactive

policy mechanism.

3.6 Network Transport Layer

This layer concerns functions that would normally be

assigned to the network and transport layers of the

OSI model. Hence this layer examines addressing,

routing and transport issues in peripheral networks.

The current opinion is that all networks whether in

the core or on the periphery should be using TCP/IP.

This thinking has been reinforced by End-to-End ar-

guments which have been used throughout the ar-

chitectural discussions when the Internet was de-

signed (Saltzer et al., 1984). The current evolution of

the Internet questions some of these End-to-End ar-

guments. As indicated previously, today the Internet

is evolving into a very fast core network with mobile

networks on the periphery. This means that charac-

teristics of the core and the periphery are diverging in

terms of latency, throughput and error proﬁles.

In the light of this, the assumption that TCP/IP

should be used in peripheral networks for heteroge-

neous networking needs to be carefully re-examined.

Firstly we should question whether IP should be used

in such networks. An assumption of the current IP in-

frastructure is that every machine should have a glob-

ally unique IP address to use on the network. This

has, however, been directly challenged by the success

of Network Address Translation (NAT) techniques. In

NAT, a private IP address space is employed in the pe-

ripheral network while only a few global IP addresses

are used to actually communicate with other machines

on the Internet. The NAT software then provides the

translation between the global IP connection and the

local machine with its private IP address. Because

all datagrams must traverse the server performing the

NAT, it provides a point in the network where incom-

ing packets can be analysed and ﬁltered as necessary.

In addition, it increases security by not making lo-

cal machines visible on the Internet, thus reducing the

potential for targeted security ﬂaw exploits, and DoS

attacks against speciﬁc machines.

The success of NAT – which is considered to be far

from ideal by network purists, including the authors

– questions the assumption that all machines should

be assigned a globally unique IP address. NAT makes

the case that IP addresses should be conﬁned to mov-

ing data within the core network. In the peripheral

networks some other form of local addressing may be

used with the translation between the networks tak-

ing place at the local gateway. Such an approach,

if deployed, will question the (often challenged) as-

sumption that there are insufﬁcient IPv4 addresses.

Though the authors support the deployment of IPv6,

the key requirement for its deployment – to provide

an inﬁnitely large global address space – needs to be

re-examined in the light of new realities.

3.6.1 TCP – Found Wanting

It is clear that TCP is unsuitable for wireless net-

works (Meyer, 1999)(Xylomenos et al., 2001). This

is primarily due to the fact that TCP interprets packet

loss as exclusively due to congestion and reacts by

substantially decreasing its send rate, and then em-

ploying its slow start mechanism. While such a con-

clusion may be valid for wired systems such as the

core network, peripheral wireless systems continu-

ously lose packets due to channel fading, interference,

vertical handovers, and other related effects. Most of

these transients are temporary and unrelated to net-

work congestion.

There have been several attempts to modify TCP

in the light of these effects, such as those described

in (Balakrishnan et al., 1997) and (Chandra et al.,

2003). Recently, there has been a move towards not

modifying the TCP protocol engine but making it

more responsive to temporary network outages (Scott

and Mapp, 2003). While this is useful, a clear, gen-

erally applicable, and elegant solution has not been

found.

3.7 The Case for Network Plurality

and Application Conformity

The idea that a different networking infrastructure

runs in peripheral networks brings with it many chal-

lenges. Most importantly, the ability to translate to

different naming and addressing schemes as pack-

ets are transmitted through different networks. Some

of issues are addressed in Plutarch (Crowcroft et al.,

2003) by the introduction of contexts and interstitial

functions. The key requirement is the development

of a framework where different networks interwork to

the beneﬁt of the applications running on those differ-

ent networks.

However, in order to ensure that every application

developed using TCP/IP will not be required to be

rewritten, the issue of application conformity must be

addressed. An interesting solution is to pursue the

idea of TCP/IP not only being a real protocol suite

in the core network but also a protocol interface in

peripheral networks. This means that on end user

devices in these networks, the TCP protocol engine

forms an overlay above whatever networking protocol

is actually being used on the network. Hence, appli-

cations can assume TCP behaviour whilst the actual

protocol in use need not be TCP. We believe that this

approach should allow network plurality to emerge

whilst maintaining application conformance.

Though application conformance may be necessary

in the short and medium terms, in the long term we

believe that the we need to move to a situation where

an application’s transport requirements are not speci-

ﬁed by the selection of a speciﬁc protocol, but by the

deﬁnition of a QoS vector which speciﬁes the trans-

port requirements of the application, as suggested

in (Mapp and Hodges, 1997).

3.8 Quality of Service Layer

Because the hetnet device will be using multiple net-

works, it will inevitably experience different qualities

of service (Rodriguez et al., 2004). Therefore, a QoS

layer is required to manage this variability. To cover

this appropriately we now introduce the concepts of

downward and upward QoS.

3.8.1 Downward QoS

The intention is to construct a series of mechanisms

to handle the different qualities of service that are en-

countered in vertical handovers in heterogeneous net-

works. In order achieve this a mechanism is needed

that bundles connections over the different available

channels, which themselves may be varying. We term

this Downward QoS, which will be required to sup-

port legacy applications.

3.8.2 Upward QoS

It is hoped that future applications will be able to

react to QoS changes as the hetnet device changes

its location. These changes will be provided via the

QoS layer. This is termed Upward QoS. In this

case, an event-based QoS signalling mechanism can

be used to inform applications of QoS changes. Ap-

plications, when they begin executing, would be able

specify routines that should be called in response to

event notiﬁcations by the QoS layer. This is sim-

ilar to the X Windows System (Scheiﬂer and Get-

tys, 1986) in which clients of the X-Server can spec-

ify what routines should be called on being notiﬁed

about certain events concerning windows they have

created on the screen. In the networking community,

the TRIGTRAN (Dawkins, 2003) and PILC (Karn,

2003) projects illustrate this paradigm, by lower lay-

ers providing hints to higher layers.

3.8.3 QoS-Aware Middleware

A hugely useful component in the support of Upward

QoS is the development of a QoS-aware middleware

platform for Distributed Environments. The concept

is to take a well known environment such as CORBA

and add support for Upward QoS capability natively

to the architecture. This would allow distributed ap-

plications to work seamlessly over heterogeneous mo-

bile networks.

3.9 The Application Environments

Layer – A Toolkit Approach

Some application environments attempt to encapsu-

late several architectural layers (Niebert et al., 2004).

These systems are therefore so large that compatibil-

ity with other systems is not really considered impor-

tant. This is an issue if we desire true connectivity

between all devices and applications.

An alternative approach is to adopt a Toolkit phi-

losophy in which the principal goal is not to build a

particular application environment but to specify the

components of a toolkit which would aid different

groups in building application environments. Hence

there would be a degree of compatibility which would

also encapsulate the functionality of the lower layers

of the architecture.

4 ARCHITECTURE TESTBED

The original ideas expressed in the framework archi-

tecture have been explored by the development of

a testbed in the University of Cambridge Computer

Laboratory. The testbed is depicted in Figure 2. Un-

like many other testbeds, our setup makes use of con-

nections to commercial providers’ networks, allowing

us to evaluate the performance of Mobile IPv6 in re-

alistic scenarios, i.e. production environments. The

architecture of the testbed is outlined in greater detail

in (Vidales et al., 2005b).

This advantage has enabled us to carry out re-

search into both the performance of cellular net-

works for data transfer (Chakravorty et al., 2002),

and also much work on the optimisation of verti-

cal handovers using client-based techniques, e.g. by

RA caching (Vidales et al., 2003) and BU simulcast-

ing (Chakravorty et al., 2004), as well as soft han-

dovers.

Recent work has focused on the latency of

TCP’s adaptation to new networks on vertical han-

dover (Cottingham and Vidales, 2005) and policy sys-

tems that sense the environment and take handover

decisions based on such data. Future work will con-

centrate on the integration of new and emerging tech-

nologies, such as UMTS (3G) and WiMax.

In terms of policy management strategies for ver-

tical handover, the original work implemented a re-

active policy mechanism called PROTON (Vidales

et al., 2004) which was written in PONDER, a policy

speciﬁcation language. PROTON was implemented

as a three-level subsystem based on a handover exe-

cution layer, which performed the handover. A pol-

icy layer allowed policy rules to be speciﬁed and was

the mechanism that reached a decision to implement

handover. Finally PROTON posessed an input/output

layer which captured relevant events and triggers and

fed them into the policy layer.

The computational requirements of PROTON ex-

ceeded that of ordinary mobile devices, so an auto-

nomic system was developed where PROTON ran on

machines in the network but downloaded a ﬁnite-state

machine to the mobile node which encapsulated these

policies in a resource-efﬁcient fashion.

5 CURRENT WORK

At present, work is being done to develop proactive

policy mechanisms. At the University of Cambridge

coverage maps of WLAN, GPRS and 3G networks

within the City are being built. The idea is to be able

to ascertain the coverage expected at particular loca-

tions. This work will then be used to develop algo-

rithms that allow users to specify a constant or min-

imum bandwidth they wish to be provided with on a

journey, for which the system will recommend an ap-

propriate route.

A proactive system based on mathematical mod-

elling is being pursued by the Networking Research

Group at Middlesex University (Shaikh et al., 2006).

This looks at simple ways of developing mathematical

models of Time Before Vertical Handover for upward

handover scenarios where a WLAN network is cur-

rently in range but will be out of range based on the

velocity and trajectory of the mobile node, whereupon

the mobile node will then perform a vertical handover

to a UMTS system. The model introduces the concept

of a Boundary Base Station (BBS) and determines the

TBVH for the node leaving the wireless network at

that BBS. The calculation of TBVH is done using pa-

rameters that can be easily determined by current net-

work deployments.

Work has also begun on examining End-to-End

Transport issues. The approach taken is to ﬁrst look at

developing a ﬂexible method of specifying networks,

and the need for certain deﬁned characteristics such

as addressing and naming. Work is being carried out

to develop a precise deﬁnition of a context as well as

interstitial functions (Crowcroft et al., 2003).

In terms of the QoS layer, a mechanism to exploit

downward QoS has already been speciﬁed. This in-

volves the development of a Stream Bundle Manage-

ment or SBM Layer (Shaikh et al., 2005). This layer

attempts to map application QoS requirements to the

available channels as the policy management layer in-

dicates the availability of certain networks as well as

the TBVH parameter. The major component of the

SBM layer is a resolution matrix in which for each

network there is a unique network id, a status indi-

cator showing off or on, the average available band-

width, the received signal strength, the time before

vertical handover and the round trip time. The resolu-

tion matrix is used to map application trafﬁc manage-

ment requirements onto available network channels.

6 FUTURE WORK

Future work involves investigating the implemention

of the upper levels of the architecture. We would

like to explore issues in the upward QoS approach,

examining how QoS events are generated from the

lower layers. These efforts also need to be compat-

ible with what is being done in the TRIGTRAN and

PILC working groups.

In addition the use of these mechanisms to build

a QoS-aware Middleware will also need to be ex-

plored. We are interested in starting with a version

of omniORB

, attempting to add the events mecha-

nism into the CORBA architecture. In terms of appli-

cation environments, we intend to port the testbed to

a commonly used mobile operating environment such

as Microsoft Windows Mobile 5 or the Symbian OS.

Finally there is a need to devise and implement a se-

curity plane for this architecture which encompasses

many layers of the framework.

7 CONCLUSION

This paper has proposed a new framework for het-

erogeneous networking. The framework allows much

http://omniorb.sourceforge.net

Correspondent Node

BSC CGSN

edge router

Mobile Nodes

IPv6−LAN Foreign Network

Default Gateway IPv6−LAN

Foreign Network

Vodafone’s live GPRS Network

WLAN Foreign Network

BTExact IPv6−Net

6BONE

WLAN Foreign Network

Home Agent

WLAN Home Network

LCE IPv4−LAN

LAN

WLAN

GPRS

Figure 2: Structure of the Cambridge Wireless MIPv6 Testbed.

greater ﬂexibility than previous models, whilst main-

taining a clear separation between the different lay-

ers. It is fully recognised that there is still much to

do. Hence the authors would welcome feedback on

the contents of this paper.

REFERENCES

Aust, S., Fikouras, N., Protel, D., Gorg, C., and Pampu, C.

(2003). Policy based mobile ip handoff decision (poli-

mand). Technical Report draft-iponair-dna-polimand-

00.txt, Work in Progress, IETF.

Balakrishnan, H., Padmanabhan, V. N., Seshan, S., and

Katz, R. (1997). A comparison of mechanisms

for improving TCP performance over wireless links.

IEEE/ACM Trans. On Networking, 5(6):756–769.

Chakravorty, R., Cartwright, J., and Pratt, I. (2002). Prac-

tical experience with TCP over GPRS. In Proc. IEEE

GLOBECOMM.

Chakravorty, R., Vidales, P., Subramanian, K., Pratt, I., and

Crowcroft, J. (2004). Performance issues with verti-

cal handovers – experiences from GPRS cellular and

WLAN hot-spots integration. In Proc. IEEE PerCom.

Chandra, D., Harris, R., and Shenoy, N. (2003). TCP per-

formance for future IP-based wireless networks. In

Proc. IASTED Wireless and Optical Communications

Conference.

Cottingham, D. N. and Vidales, P. (2005). Is latency the

real enemy in next generation networks? In Proc.

ICST CoNWiN.

Crowcroft, J., Hand, S., Mortier, R., Roscoe, T., and

Warﬁeld, A. (2003). Plutarch: an argument for net-

work pluralism. SIGCOMM Comput. Commun. Rev.,

33(4):258–266.

Dawkins, S. (2003). End-to-end, implicit ’link-up’ no-

tiﬁcation. Technical Report draft-dawkins-trigtran-

linkup01 (work in progress), IETF.

Karn, P. (2003). Advice for internet subnetwork designers.

Technical Report draft-ietf-pilc-link-design-15 (work

in progress), IETF.

Laasonen, K., Raento, M., and Toivonen, H. (2004). Adap-

tive on-device location recognition. LNCS 3001,

pages 288–304.

Mapp, G. and Hodges, S. J. (1997). QoS-based transport.

In Proc. IFIP Workshop on Quality of Service.

McNair, J. and Zhu, F. (2004). Vertical Handoffs in Fourth-

Generation Multinetwork Environments. IEEE Wire-

less Communications, 11(5).

Meyer, M. (1999). TCP performance over GPRS. In Proc.

IEEE WCNC, pages 1242–1252.

Niebert, N., Schieder, A., Abramowicz, H., Malmgren, G.,

Sachs, J., Horn, U., Prehofer, C., and Karl, H. (2004).

Ambient Networks: An Architecture for Communica-

tion Networks Beyond 3G. IEEE Wireless Communi-

cations, 11(2).

Rodriguez, P., Chakravorty, R., Chesterﬁeld, J., Pratt, I., and

Banerjee, S. (2004). MAR: A Commuter Router In-

frastructure for the Mobile Internet. In Proceedings

of the ACM Second Mobile Systems, Applications and

Services Conferences (ACM Mobisys 2004).

Saltzer, J. H., Reed, D. P., and Clark, D. D. (1984). End-to-

end arguments in system design. ACM Trans. Comput.

Syst., 2(4):277–288.

Scheiﬂer, R. W. and Gettys, J. (1986). The X window sys-

tem. ACM Trans. Graph., 5(2):79–109.

Scott, J. and Mapp, G. E. (2003). Link layer-based TCP

optimisation for disconnecting networks. Computer

Communications Review, 33(5).

Shaikh, F., Lasebae, A., and Mapp, G. (2005). SBM layer

for optimum management of co-existing telemedicine

trafﬁc streams under varying channel conditions in

heterogeneous networks. In Proc. MIT.

Shaikh, F., Lasebae, A., and Mapp, G. (2006). Client-based

SBM layer for predictive management of trafﬁc ﬂows

in heterogeneous networks. In Proc. IEEE ICTTA.

Soh, W.-S. and Kim, H. (2004). Dynamic bandwidth reser-

vation in cellular networks using road topology based

mobility predictions. In Proc. IEEE INFOCOM.

Vidales, P., Baliosian, J., Serrat, J., Mapp, G., Stajano, F.,

and Hopper, A. (2005a). Autonomic system for mobil-

ity support in 4G networks. IEEE Journal on Selected

Areas in Communications.

Vidales, P., Chakravorty, R., and Policroniades, C. (2004).

PROTON: A policy-based solution for future 4G de-

vices. In Proc. IEEE International Workshop on Poli-

cies for Distributed Systems and Networks.

Vidales, P., Mapp, G., Stajano, F., Crowcroft, J., and

Bernardos, C. J. (2005b). A practical approach for 4g

systems: Deployment of overlay networks. In Proc.

TRIDENTCOM, pages 172–181.

Vidales, P., Patanapongpibul, L., and Chakravorty, R.

(2003). Ubiquitous networking in heterogeneous en-

vironments. In Proc. 8th International Workshop on

Mobile Multimedia Communications.

Xylomenos, G., Polyzos, G. C., M

ahonen, P., and Saara-

nen, M. (2001). TCP performance issues over wireless

links. IEEE Communications Magazine, 39(4):52–58.

Zimmermann, H. (1988). OSI reference model – the ISO

model of architecture for open systems interconnec-

tion. Innovations in Internetworking, pages 2–9.