Evolution Toward 4G Communication System

Claudio E. Palazzi

Department of Computer Science, University of Bologna Mura A. Zamboni 7, 40127 Bologna, Italy

E-mail: cpalazzi@cs.unibo.it

Advisor: Prof. Marco Roccetti

3

1 - Introduction The current status of the Internet as a widely used tool and the overwhelming

development and popularity of wireless access technology lead us to a future in which the synergy between wireless and the Internet will be an integral part of our everyday life. Virtual libraries, remote-working, video-telephony or voice over IP, on-line games, traffic control, remote-medicine, video and music on demand, location based resource discovery, navigation support, are only a few of the innumerable services that will be available in every place and at every time [112][113][114]. People will be allowed to be continuously connected during the whole day, regardless of their location (eg, home, workplace, car, airport, hotel, etc) and utilizing a plethora of traditional or new devices (eg, PCs, laptops, PDAs, cell phones, other handheld device, next generation appliances, etc).

We are crossing a technology threshold that will revolutionize every area of our

lives. It will affect all of our everyday habits and businesses in ways far more pervasive than most people may imagine. Devices that we use today for a limited range of special purposes will become multiple application platforms. Even common objects as wristwatches, cars, PDAs are evolving and their enhancement toward multipurpose tools will accelerate as we move forward. Wristwatch capabilities could be augmented making it able to communicate, download/play music, keep personal/medical information, identify us to our car/home/devices, etc. Cars will be elevated from a simple transportation vehicle to an office on the move, as well as an information provider and entertainment center. Passengers will be allowed to access the Internet, engage in teleconferencing, play distributed videogames, learn location based information as low traffic paths to the destination or special offers for hotel reservations, participate in ah hoc or mesh networks [81], etc. Paper money and coins are going to be completely substituted by electronic transactions. Credit cards already are more and more frequently used by customers for their real or on-line purchases. The availability of always connected devices could further push commerce in this direction. Pocket sized PDAs, in fact, could be enhanced to be an easy-to-use way of payment. Customers will buy their objects and services connecting to a web page or simply interacting their PDA with the cashier. Even tickets for events or travel will disappear in the paper format: PDAs or equivalent multipurpose devices will be able to provide the information required to authenticate the completed transaction.

In the aim of customers’ fully satisfying experience several issues have to be

guaranteed. A secure and efficient connectivity have to be always available even in case of seamless switching between access network technologies. Moreover, it is crucial that all the devices could interact one to each other to continue and successfully complete a session started from another terminal. In order to clarify our vision of the future with some examples let’s imagine a student using his/her DSL connection for streaming a compilation of music on-demand [95] on his/her connected Hi-Fi at home. At a certain time the calendar function of his/her wristwatch remind him/her an appointment with a friend in a cafeteria. Before leaving home, he/she redirects eventual calls to his/her Video-telephony enabled PDA. After this, he/she transfers the music stream on his/her PDA, automatically connected with the home Wireless Local Area Network (WLAN), and then exits. Once the terminal is inside the car and as far the car moves away from home, the WLAN signal becomes

4

weaker and weaker and the device autonomously discovers and selects a better network access, thus switching to UMTS [86]. The student continues enjoying the selected songs, now directly from the high quality automobile speakers. Once selected the destination, the on board computer in the car sets the most appropriate route until a mutated traffic congestion information is received through the urban traffic grid [83]; the navigation system suggests a deviation and, finally, a parking lot close to the cafeteria. Sensors in the parking space detect that specific car by its unique identifier chip and automatically charge the owner for the utilized time. Music stream is again transferred to the PDA while the student leaves the automobile and starts hearing the compilation from his/her Bluetooth [110] enabled earphones. Entering the local, the Always Best Connected [13] system of the PDA automatically detects the presence of the free WLAN of the cafeteria and switches to it. The student meets his/her friend at a table and suspends the stream: he will be charged proportionally to the “consumed” service. His/her friend looks at the multimedia player and suggests to immediately download from the P2P network a new freely available enhanced software for multimedia entertainment…

We are leaving in a Communication Era where computers and connectivity are

becoming increasingly personal and essential. In this scenario, imagine also an emergency situation when power goes out and all the Hot Spots and Cellular Base Stations shut off thus impeding any kind of communications. Similar crisis conditions in an urban area could occur when there is a chemical or nuclear disaster caused by human error, plant break down, act of war or terrorist attack. Yet, this is the time when communications are indispensable to control all the emergency operations and a rapid deployable connectivity must be guaranteed. In these circumstances, satellites combined with the use of sky objects flying above the crisis area, could represent the only communication way left to the terrestrial infrastructure.

While the successful development of 2G technology was determined by a

combination of worldwide raising economical trend, a “killer application” as the Short Message Services (SMS) and the possibility to use pre-paid cards to recharge the credit, the same starting conditions lack in 3G, whose deployment has been further delayed by the very expensive and time-consuming spectrum license auctions. This situation coupled with the increasing popularity of the unlicensed spectrum family of IEEE 802.11 standards [67] has been part of the impressive growth in the deployment of WLAN access hot spots [85]. The high speed data service provided by WLAN and its minimum initial investment and operational cost make this access technology a natural co-player for the 3G systems: applications run indoor by 3G users could in fact take benefit from being able to exploit the higher rate offered. Moreover, access in hot spots could be free or flat rate for the user, thus giving another important reason for switching type of connection while entering a WLAN.

The above depicted situation leads us toward an Always-On future. Not only,

people are going to demand the technology required to be Always Best Connected. This new philosophy will permit users to exploit the locally best offered access for their needs. In each moment and ubiquitously, a person will be able to chose between the various connections available (wired access, Bluetooth, WLAN, 2.5G, 3G, etc) or a combination of them, and between the various devices available depending on the application requirements (eg, screen size, energy consumption, mobility, processing capabilities, network interfaces, etc). Devices should be configurable such that usera

5

only need to set some initial preferences in order to have, at any point in time, an automatic and almost invisible best access choice. Indeed, communications utilizing an IP-core network will be able to guarantee connectivity even between diverse access technologies. Therefore, the 3G following generation of communication, namely 4G, will not necessarily rely on an independent new radio infrastructure, instead its peculiarity will be the integration of heterogeneous segments exploiting different technologies within the common glue of the Internet.

In the following section we present a more precise description of the various

problems that arise in this scenario, while in section 3 we list the current state of the art in the various technologies involved. In section 4 and 5 we show respectively the work we have done till now and the results attained. We conclude with future directions of this work for the next years in section 6.

7

2 – Problem statement An Always Best Connected scenario involves several levels of complexity that

spread both on technological solutions for terminals and networks, and on business agreements between access operators, service providers and final users. In this work, we will focus on the technical part leaving the economical issues to experts in that field.

A mobile connection to the Internet from heterogeneous devices and utilizing

multiple access technologies needs robust support in order to offer a pleasant experience. The considered environment, in fact, is complicated by several factors:

users mobility with consequent roaming, fading, disconnections, and latency

variability; heterogeneous devices with different features and constraints; multiple wireless access technologies; users interaction and preferences in independently choosing access scheme; time varying traffic pattern (by load, type and priorities); system prone to high interference and congestion (elevate packet loss ratio and

highly variable load); If we include in the depicted scenario also the need for an Always-On connectivity,

even in case of urban disaster, the main designing challenges that we have to face can be summarized as follows:

a) emergency architecture involving flying objects and satellites; b) wired-cum-wireless protocols for efficient and fair use of the channel; c) horizontal/vertical seamless handoff capabilities; d) rate/content adaptation depending on network access or device utilized; e) QoS maintenance or re-negotiation through variable environment or access

technology; f) session mobility between heterogeneous terminals; g) wireless security; h) access discovery/selection mechanisms; i) wireless GRID and Peer-to-Peer extensions.

a) In an urban crisis scenario, where some disaster has brought power and telecommunication systems to a standstill, it is very cost effective to deploy in a short time a system composed of several HAPS (High Altitude Platforms Stations) or UAVs (Unmanned Airborne Vehicles) [6][7][8] to establish an emergency telecommunication infrastructure. The HAPS/UAV may fly through the “urban canyons” acting as repeaters. The mobile users on the ground can use HAPS/UAVs to communicate with each other and to access a remote ground station or the Internet via Satellite as depicted in Fig. 2.1. This two level “satellite empowered” architecture combines the advantage of having a very small user terminal technology on the mobiles with the capability to establish very long range connections. The HAPS/UAVs, can be rapidly deployed, thus providing communications within minutes from the accident. Once in place, the HAPS/UAV will act like a Hot Spot to the customers on the ground. Thus, the access protocol will be 802.11, compatible

8

with existing Hot Spot environments. The effectiveness of this architecture coupled with an enhanced wired-cum-wireless designed transport protocol and the eventual split of the connection at the HAPS/UAV have been analyzed in sections 4.1 and 5.1.

Gateway

802.11

GEO Satellite

UAV + proxy

Urban Environment

Fig. 2.1 – HAPS/UAV and satellite communication scenario

b) Since the TCP/IP suit of protocols was designed in a time when networks were based exclusively on wired technology, its mechanism fails when plunged into a wireless environment. Not only a connection with free mobility requires new routing discovery schemes, but also the flow control and congestion control functions of TCP results heavily affected by the high loss ratio that characterizes wireless transmissions [14][72]. In fact, TCP uses packet losses as a metric to evaluate the congestion level of the network, thus shrinking the sending window. In presence of numerous losses related to non-congestion factors, as in a wireless environment, this behavior is not appropriate and causes a consistent underutilization of the link bandwidth [17]. The 802.11 MAC layer protocol attempts to face the packet loss problem by implementing its own retransmission scheme [16]. This scheme hides wireless error losses from the TCP’s congestion control mechanism, thus avoiding deleterious multiple reductions of the data sending window [100][103][104]. On the other hand, local retransmissions affect packet delivery delay by increasing its variability and thereby particularly impacting on time-constrained applications such as audio or video stream. c) In a mobile scenario, horizontal handoffs occurs regularly, every time the user crosses two cells of coverage for the technology in use [64]. Moreover, as depicted in Fig. 2.2, the notion of handoff as the last resort to maintain the connectivity needs to be enhanced with the concept of vertical handoffs [66]. When a mobile is entering an area where the currently used access technology cannot anymore be conserved, or simply there is a better connectivity option, the device should automatically and almost invisibly switch connection. The transition should be smooth in order to preserve a regular communication and avoid performance degradation that could annoy the customer’s activities. This raises several problems, among the others, the handoff procedure should take into account the diverse nature of the underlying technologies and the timing of the handoff should be optimized taking in consideration a multitude of parameters like network congestion, link quality, energy consumption, pricing, user’s preferences, etc.

9

Fig. 2.2 – Horizontal and Vertical Handoff

d) In the depicted scenario, a multitude of customers could use ubiquitously the same application from various hardware equipments. Not only, in every moment a user could need to switch device type or access technology without interrupting the on-going session. This two levels of complexity are not well supported by the current Internet infrastructure nor by the application capabilities. The deployment of agents, proxies and “smart” clients could help in taking intelligent decisions and sharing responsibilities regarding caching and pre-fetching, data delivery ratio, compression level utilized, etc. This combination of efforts have to rely on scalable adaptation techniques and encoded content suited for on-the-fly adaptation [106]. Moreover, since the diversity of hardware and access technology employable, decision about content adaptation should regard also an efficient tradeoff between CPU cycles and energy consumed in using another encoding for the data versus bandwidth wasted in the delivery of extra information in each packets. e) In a scenario where users dynamically changes the best access to multimedia services, the presence of Quality of Service (QoS) support is crucial. However, a mere extension of QoS schemes utilized in wired connections can hardly accomplish the requirements of a wireless environment made even more complex by horizontal and vertical handoffs. The next-generation wireless multimedia communication systems, in fact, requires efficient mechanisms for QoS provision able to operate even over unreliable, unstable and mostly unpredictable channels [75]. Even in those conditions, the QoS system have to be able to assign Traffic Categories (TC) to the various flows and to enforce Content Based Policies (CBP) within the same flow depending on the different level of importance of each packet with respect of the final perceived quality by the user. f) The Always Best Connected philosophy passes through the utilization of different devices characterized by their own features and, among these, the user can chose the one that best suit his/her needs. Furthermore, at every time, a person should be allowed to change the hardware means utilized to run the ongoing applications without losing the work done till that moment. In practice, platforms have to be able to seamlessly transfer their active sessions between them every time the user decide it. Not only this migration has to avoid loss of consistency, it also have to be as fast as possible (if not instantaneous). Specifically, if the running application is a stream of some kind of data, then the we only need to move some connectivity information; otherwise, in case of downloading, the amount of data to be transferred could be greater unless having previously stored part or all of it in both the devices.

10

g) A ubiquitous use of mobile terminals across a variety of wired and wireless connections requires an effective security architecture. Not only nowadays people are feeling privacy as a fundamental right but security is an indispensable component in every transaction. Even with various advancements in many issues related to wireless technology, security requirements are still difficult to guarantee [71]. The open nature of mobile access, with communication signals that spread from the source in every direction around it, represents a very tough obstacle to security. Despite of this, in order to really reach customer’s needs, wireless technology have to be enhanced to authenticate and authorize users utilizing system resources. The integrity of the data must be preserved and authorized users have to be the exclusive subjects able to receive and manage their specific data contents. Finally, both these requirements have to be guaranteed even during seamless session handoffs. h) The mobile terminal has to be enhanced with an access discovery function that periodically looks for better alternatives. In order to classify the diverse access options and being able to efficiently choose between them, we need to find the best tradeoff between several issues. First of all, devices have to be able to discriminate between access network technologies, operators, pricing, QoS required and user’s preferences through a set of general parameters. Then, we have to define a set of metrics on the various access types that can be used as statistics to determine the best choice for future accesses. The selection process is made of choices that are based on the terminal used, the networks available and user’s predilections. The latter part requires the presence of an easy-to-use tool that shows the various options in a way that supports customers in taking decision that are beneficial for his/her needs. Finally, in both cases of first access and loss of currently used connectivity, the terminal must be able to discover and select another access option without any support from the network. i) Wireless communications, Peer-to-Peer (P2P) and GRID networking probably constitutes the three highest scale technology trends of the past few years. Grid and P2P have much in common: both are concerned with the coordinated use of resources within distributed communities and operates as overlay networking structures. Focusing on P2P, its name has been closely associated with Napster [88], which captured the media attention with the forty million downloads of its music-sharing software and the copyright issues involved in the phenomenon. Wireless and P2P are naturally linked; the increased capabilities of wireless devices, in fact, will make them people’s favorite media-players, thus having in a P2P file-sharing service a standard feature. Moreover, information gathering involving physical proximity to a specific location could receive more efficient service utilizing direct connections instead of a centralized architecture. At the same time, GRID technology has emerged as the ideal tool to create scalable virtual organization of computers [84]. Its capability would be especially useful if we imagine a scenario requiring more powerful resources than those available in a mobile terminal, as it could be an urban traffic grid. In this case, terminals installed in cars could contribute in creating the general traffic situation with their location based information, but it will be up to a central main frame to calculate and distribute in real time the information about the less congested routes to drivers.

11

3 – Related work

Many research works in the past years have focused on issues related to a connection from a single device type through a unique network interface. Wireless accesses have been studied from several standpoints: handoffs, performance, new protocols, QoS maintenance, security, etc. However, the scientific literature still has not comprehensively addressed the area of multiple wireless interfaces. The MosquitoNet project [18][70] falls in this area but it mostly tackles IP issues just mentioning, without providing any factual algorithm, eventual QoS support by the use of multiple interfaces. Also, this project does not delve into power issues arising from using multiple interfaces.

In literature, we can find many examples of wireless dedicated transport protocols

[15]. The inter-arrival times between data packets is used in [19] as an interpretative metric for discriminating between congestion and error losses. In order to face problems related to disconnections and handoffs, [20] proposes to use the standard TCP function of sending a zero advertised window to the source: if received in time, this will suspend transmissions and thus losses till a full connectivity restore. Special probing packets are used in [21] after each loss to determine the condition of the channel: only if their delay indicates congestion then the sending rate is decreased. In case of poor channel conditions one of the two probing packets gets probably lost, the probing cycle is thus extended with an exponentially increasing departing interval. Since the small dimension of probing packets, this scheme also saves energy in case of temporary disconnections. A new end-to-end transport protocol is also proposed in [25]. This work faces the high number of errors and the variable latency of a wireless environment eliminating the timeout mechanism, employing periodic SACK packets [26] to understand when a retransmission is appropriate, and estimating the channel capacity to set the data sending rate. The key innovative idea behind [22] is the use of the data acknowledged in the time unit in order to sender-side measure the effective bandwidth available on the link. This bandwidth estimation is computed by sampling and exponential filtering methods and then used, after a loss or during slow start, to appropriately set the slow start threshold.

Much work has also been done on Power Saving Mechanisms [27][28][29][30] for

the single interface. Some specific examples are the work on the Bounded Power Save Mode (BPSM) [31] and on measurement of power consumption of small devices [32]. When using multiple interfaces like 802.11 and Bluetooth, interference issues comes from the utilization of the same frequency band. The interference between 802.11 and Bluetooth has been studied in [33][34] and ways to address it have been proposed in [35]. On the other hand, how each interface should be turned on/off to minimize this problem when these interfaces are being used on the same terminal has not been studied yet. However the low power modes of Bluetooth [36] could be very useful in implementing such power-save and interference-aware algorithms.

The handoff concept started with connections in cellular networks and regards a terminal moving from one cell to another. This generic situation of horizontal handoff between networks using the same access technology, was explored also in the 802.11 domain. There have been numerous papers dealing with horizontal handoffs across homogeneous cellular [44], ATM [45], picocellular [46] networks and mobility in IP

12

networks [47]. We talk about vertical handoffs if we focus on switching connectivity between networks that utilize different access technology. The BARWAN project [48] has done a seminal work in this area considering also various access technologies: infrared, Wavelan and Richocet and other proposed techniques to reduce handoff latency across technologies. A session layer is introduced in [82] to effectively address disconnections due to horizontal and vertical handoffs during seamless music delivery service to mobile users. The work presented in [49] includes some measurements to calculate the amount of power consumed by the different interfaces and looks at application oriented approaches to save power. Though this work explored the plethora of issues related to heterogeneous access technologies, there still are a lot of open problems some of which have arisen later due to the appearance of technologies such as Bluetooth, 2.5G and 3G. Vertical handoffs in those technologies lead us to the core of this work: the evolution toward the open field of 4G communication system.

IP telephony has been investigated by the scientific community by a decade and

has today the status of a commercial technology [37]. Research efforts have been put in the area of buffer management in order to allow an effective voice transmission even in a jitter affected network [38]. The increasing success of WLAN and the emerging needs for ubiquitous computing have triggered several studies on the VoIP performances in IEEE802.11 networks such as [39][40][41][42]. Finally, mobile extensions for the IP telephony have been proposed but the seamless vertical handoff of a phone call is still an open issues [43].

The attempt to guarantee the customer with a pleasant connection experience passes through having efficient and scalable content adaptation and delivery in wireless networks. This has generated a number of solutions relying on transcoding of data [50][51][52][53][54]. Depending on channel conditions, data are sent with different compression/quality level, thus limiting byte transmitted when necessary at the cost of some additional computational work.

Adapting contents to channel condition can be done also by using proxies [58].

Indeed, many solutions have advocated the use of proxies especially to improve the performance in wireless networks [55][56][111]. We present a possible employment of this technique in sections 4.1 and 5.1. Furthermore, [57] studied the behavior of HTTP and TCP in GPRS networks, their impact on browsing the web in such networks and the utilization of proxies to alleviate this problem. It has also been proposed that proxies be used to perform caching and/or prefetching in wireless networks [60][61][62].

Enhanced proxies could also helps in enforcing QoS requirements in wireless

networks [59]. However, the transfer of QoS information, especially when handoffs occur between dissimilar networks has not been studied in great detail. Indeed, the status of the art of the proposed mechanisms allows to provide QoS differentiation but no guarantees of QoS levels [74]. The emerging idea in this field is to replace the hard QoS guarantee utilized for wired connection with a soft QoS scheme [73]. Focusing on the QoS support in the currently leading WLAN standard, we notice that the IEEE802.11a, b, and g, version do not have any QoS management unless extended with the IEEE802.11e. Moreover, the QoS support in the IEEE802.11e still presents several open issues to be addressed.

13

Session mobility between heterogeneous terminals is a relatively new topic and

techniques to address it comes from still ongoing works as the iMASH project [68][69]. The scalability of the mechanisms presented relies on a Distributed Middleware Service layer that maintains application session handoffs consistency even in case of a large number of client [76]. Even if the specific target of this project is healthcare applications, the proposed solutions can be extended to a general mobile application. Another possible technique utilizes Virtual Socket [77], on top of the real ones, in order to resume transmissions after a session handoff .

Much work has been done even in the field of wireless security. In order to provide

authenticated end-to-end security, the already cited iMASH project [78] employs various techniques as the Bell-LaPadula models coupled with Public Key Infrastructure. In [79] symmetric key solutions results equivalently secure and less costly than public ones in a wireless environment. However, the use of a particular set of protocols for the test doesn’t allow to generalize with certainty this assertion. A security protocol specifically designed for a WLAN environment is presented in [80]. The exploitation of tokens, in fact, allows to utilize a simple security scheme, avoiding too sophisticated and resource-consuming cryptographic techniques, thus taking into account characteristics of the wireless media as the limited bandwidth and computational power. Finally, the security protocols for the IEEE 802.11b and the implications for user privacy and utilization are given in [89].

Both P2P and Grid have raised great interest in the scientific community. A

comparative analysis between the two technologies is presented in [90]. Even with differences in final users and in computational capabilities required, the two technologies present various points of contact in their unsolved issues. Considering P2P, researches have focused on the creation of fully decentralized networks. In particular various Distributed Hash Table (DHT) frameworks have been developed to efficiently route messages [91][92][93]. However, many other challenges still need to be effectively addressed, including complete anonymity, trust between nodes, efficient content searching, etc. The Grid is defined as a “flexible, secure, coordinated resource sharing among dynamic collections of individuals, institutions, and resources—what we refer to as virtual organizations” [94]. Carefully reading this definition, we find several challenges for a factual implementation, many of which can be shared with P2P technology: discovering and utilization of remote resources, forming and information maintenance of large distributed virtual communities, preservation of personal and remote security, etc. There is no need to say that a wireless scenario, in particular mobility, further increases the complexity of the cited problems.

15

4 – Current approach

Since the differentiated nature of problems that are involved in a non homogeneous wired-cum-wireless environment, the realization of a factual Always Best Connected world cannot rely on a single technology. Indeed, various integrated approaches should be taken in mind. The final user have to be protected both from the plethora of problems characterizing the wireless means and from the hardness in choosing the best strategy in every moment. In this section we present a couple of strategy we have developed to address the former issue. The common glue between the two proposed solution is the presence of TCP Westwood: an enhanced end-to-end version of TCP able to deal more effectively on connections having a wireless segment between source and destination. As anticipated in section 3, this new transport protocol utilize a sender-side Eligible Rate Estimation to discriminate between congestion and wireless errors. After a loss experience in fact, the slow start threshold is set with the last estimated sustainable bandwidth rather than being blindly reduced to half of the congestion window. Error losses while still having a low congestion window, in fact, makes standard TCP New Reno [105] prematurely exits the slow start phase even in absence of congestion, thus substantially decreasing performance. With TCP Westwood, the slow start threshold is maintained at a correct level ensuring an efficient and non aggressive data rate [107][109]. TCP Westwood estimates its eligible rate calculating the amount of data acknowledged in a certain period of time which varies depending on the ongoing congestion sensed [23]. Last version of this transport protocol, namely AGILE [24], also incorporates a mechanism to detect non congestion situations and increase the slow start threshold to an appropriate value while still regularly transmitting. Our work shows the benefits attained by this protocol coupled with a proxy employment on a satellite link utilized to restore connectivity in an urban crisis scenario. Moreover, a capacity estimator is suggested: indeed, a possible employment of this tool could be in helping TCP Westwood to avoid bad samples while computing the shared bandwidth.

4.1 – On-board proxy technique In certain undeveloped environments or emergency urban situations, satellites may

represent the only communication way left. Moreover, we can introduce an innovative architecture involving the exploitation of unmanned flying objects (HAPS/UAV) or stationary sky station at relatively low altitudes to reduce the shadowing impairment. Splitting the connection into two parts and creating a short range link with the user terminal directly faces the very high loss ratio present especially in the first part of the connection. In fact, geographical or urban obstacles could reduce the satellite visibility and the working communication devices left on the ground could not have enough signal power to communicate directly with the satellite. The two systems can be coupled having the latter the role of collecting information from user terminals and fast recovering from error losses thanks to the very short range and good elevation angle achievable even in urban areas, while the former can provide interconnection with the terrestrial infrastructure.

Not only voice communication, but also web traffic, area maps downloading,

image files and emergency reports forwarding, and many others could be possible

16

application running in this scenario. Many of these applications run on TCP, thus it will be important to evaluate the performance of TCP in this environment. We propose to study different ways of maintaining TCP connections:

a) Maintaining an End to end connection from ground user to Internet server,

implementing a specifically wired-cum-wireless designed transport protocol: TCP Westwood.

b) Splitting the TCP connection boarding a proxy server on the HAPS/UAV.

Dividing the problem into two subproblems easier to manage allows to face in loco, and with a shortened propagation delay, the high error rates characterizing the first part of the connection.

c) Using a combination of a new transport protocol and the splitting scheme to

reach the maximum efficiency on every sublink.

The basic idea behind the so-called PEP (Performance Enhancing Proxy) [9] is the involvement of intermediary processing of the flow of data on behalf of TCP endpoints. The splitting scheme here proposed makes use of TCP gateways and maintain multiple TCP connections with both other gateways and end users. In fact between gateways, splitting may utilize a specialized or optimized transport protocol. Even if in literature we usually find splitting on-board satellites [10], the same technology can be equally effective implemented in HAPS/UAVs. In this latter case, moreover, the put in use and eventual updates can be performed much easily and much less costly.

The scheme here proposed utilizes a division of the path into uplink and downlink

and requires the presence of a cache to store packets received and not still transmitted [5]. Of course, splitting the connection violates the end-to-end paradigm, but this problem could be alleviated in two ways. First, acks are sent back from the HAPS/UAV to the source when the packet is successfully transmitted on the channel down to the mobile and not as soon as it receives the packet. This decreases the probability of having an unrecoverable situation due to packet corruption while stored in cache. Second, the signals of correctly closed connection does not follow the splitting mechanism: the signal should be received from the mobile device and then forwarded.

In order to achieve the best results, the cache should be adequately dimensioned.

However, the advertised window sent back to the source is claimed accordingly to the space left in the cache, thus avoiding packet losses due to overflooding. On-board splitting can provide disparate terminals to communicate through the forwarding agent, that is, it supports TCP terminals without requiring uniform end-to-end TCP. Moreover, some hops may use multicast, and thus more effectively exploit the broadcast nature of wireless transmission while still supporting TCP endpoints. An on-board transport agent also provides more rapid recovery from errors and improves the overall robustness of the end-to-end connection when several links suffer from shadowing or high error-rates.

17

4.2 – Residual Capacity Estimator A packet train is a set of packets which depart from the sender grouped together

with no idle time between transmissions [65]. Traveling along the path, these packets could be delayed in different ways such that the destination will observe a different distribution than the starting one. Packets, in fact, could be dispersed by narrow links or could be delayed in queue for different times if concurrent traffic is present on that link. Acks leaves the destination at a time that depends on the correspondent packets receiving time and their dispersion once reached the sender could be used in order to infer information as the capacity or the bandwidth of the bottleneck.

In a normal TCP connection, packets does not leave the source one close to the

other in a train configuration, instead, their departure time is beaten by the returning acks time. Therefore, it is not possible to use a packet train technique with TCP, unless we modify it. At the same time, sending real packet trains on the channel could suddenly increase the congestion level thus causing multiple losses and consequent performance degradation. For all these reasons, we propose a scheme named Residual Capacity Estimator (RCE) [101] that is perfectly embedded in the normal TCP behavior and, at the same time, recall the packet train technique. We claim that RCE is a very simple and cost effective mechanism able to calculate the bottleneck capacity deducted the uniformly distributed traffic present on the path. RCE can have several applications as, for instance, discarding overestimating samples in bandwidth estimation technique, determining the appropriate routes/tree for multicast overlay networks [11], calculating the most appropriate rate in video/audio streaming.

In Fig. 4.2.1 time is divided into slots and a single TCP sends a packet train toward

the receiver. We indicate with X the part of the slot used to transmit the packets back to back and with Y the time needed to contain the dispersion of the correspondent acks. Considering the time slot equal to 1, the maximum portion of the slot usable to send packets corresponds to

YX .

Fig. 4.2.1 - Acks dispersion due to the bottleneck link with a single flow present

Let’s suppose now to introduce other traffic in this scenario. In this case, we have

to take into account also the possibility that packets could leave the source distributed on the whole slot. In fact, packet departures from the sender depends on the acks arrival and both this and the factual packets delivery at destination result delayed by the presence of other flows on the path. In a general case, actually, the acks distribution is characterized by gaps that depends on various factors:

pkts

dispersed acks

SEN

DER

REC

EIVER

X

Y

time slot

18

a) Different capacity between outgoing link and bottleneck link b) Queuing time caused by congestion c) Different size, and thus channel occupancy, between data packets and acks d) Wasted time due to a low sending window

In particular, point d) regards the fact that if we have a very low sending window

compared to the effective achievable data rate, the sender will experience long inactive periods just waiting for the returning acks. Many formulas for computing the capacity or the shared bandwidth link, relies on some ratio between the bytes acked in the unit time [87][108]. If we leave this wasted time into those formulas we will have a very underestimated value as result. Having a means to take in consideration only the first three causes gives us the ability to provide a bandwidth estimator, while considering only a) and c) we could build a capacity estimator.

RCE scheme divides time both into packets-slots and acks-slots. The formers are

set as large as an RTO [12] and in that period we count packets departing from the source. When the correspondent returning acks come back we determine the size of the acks-slot as shown in Fig. 4.2.2: the slot starts with the end of the preceding one and ends with the receiving of the ack for the last packet sent in the correspondent packets_slot. Since the number of packets-slots and of acks-slots is the same, on average they will have the same length.

Fig. 4.2.2 – Packets-slots and acks-slots division When the acks corresponding to the sent packets come back, RCE computes the

bottleneck capacity as timeWastedtimeslotAcks

slotinackedBits___

___−

. The introduction of

Wasted_time in this formula, is the enhancement that allows us to keep in consideration just the elements of the acknowledgment dispersion that reflect the characteristics of the channel without being disturbed by the time wasted at sender due simply to a low sending window. For this purpose, our scheme measures the average of the interarrival time between the acks of the current acks_slot. The Wasted_time is then computed as the sum of the time exceeding this average in each interarrival time included in the acks_slot. The rationale of this formula is the fact that all the packets will experience the same channel conditions in terms of transmission

Packets departures

Corresponding acks arrivals

Effective tx time Sender waiting time

19

time, but not all of them will endure the wasted time due to a low sending window. Therefore the average exceeding gap times between acks is most likely a result of having periods of no transmissions due to the sending window size than for channel characteristics. Focusing on queuing time, we have to notice that, since the bulk nature of TCP transmission, this element will not be endured evenly by all packets in a slot. Consequently, unless the link is experiencing heavy congestion, queuing time is removed by our mechanism in case of contemporary presence of other TCP flows and the final result is the capacity of the bottleneck. Conversely, if the considered TCP connection competes for the channel with other CBR flows, the uniform data rate of these flows will equally impact on all the TCP packets. Thereby, the final estimation will count also queuing time caused by the CBR traffic thus computing the shared bandwidth. Summarizing what we demonstrate in Section 4.2, we can say that RCE returns the bottleneck capacity, detracted the portion of channel occupied by the uniformly distributed traffic. Moreover, this is obtained since the very beginning of the connection and no special packets or TCP modified behavior are required.

The pseudo-code of RCE given in Tab. 4.2.1 shows that the scalability of RCE is

assured by the very easy set of calculations and by the very few information we need to store at sender side about the TCP flow. Moreover, the proposed mechanism is perfectly embedded into the standard TCP operations.

• At sender side, time is divided into slots • In each slot, N packets are sent to destination • The corresponding N acks will return back in Acks_slot_time • Sender_waiting_time is calculated as:

Sender_waiting_time = 0; calculate AVG_acks_interarrival_time; for each acks_interarrival_time of the slot {

if acks_interarrival_time > AVG_acks_interarrival_time { Diff = acks_interarrival_time – AVG_acks_interarrival_time; Sender_waiting_time = Sender_waiting_time + Diff;

}; };

• The sample is: Cap_sample = Bits_acked / (Acks_slot_time – Sender_waiting_time)

• This is averaged as: Cap_est = 0.5 * Cap_sample + 0.5 * Cap_est

Tab. 4.2.1 - The proposed scheme

21

5 – Current results In order to test the efficiency of our schemes, we have run several simulations

using the popular NS-2 network simulator [1]. The reliability of its outcomes and the validity of its models, especially in new transport protocols investigation [3], have driven our choice toward this platform. For results presented in section 5.1 we have adopted version 2.1b8a of this software coupled with some new modules to implement the proxy behavior. In section 5.2 we have preferred version 2.1b7a because of the availability of the NOAH extension which allows sending MAC frames directly to a specific mobile node without any routing interference in the results [2]. Moreover, in all the version utilized we have added the code to simulate the behavior of TCP Westwood. The following results and charts are taken from our works [5] and [101].

5.1 – On-board proxy utilization gain Several runs of simulations have been averaged in order to verify the gain in using

an on-board proxy in a scenario with HAPS/UAV connected to the satellite. The simulated topology is simply represented in Fig. 5.1.1 where delays and link capacities have been chosen coherently with the environment depicted in section 4.1 and Fig. 2.1. The packet size utilized is 1500 Bytes, the buffer available between W and U is 50 packets and the cache on the proxy, when splitting mechanism is enabled, amounts to 200 packets. The FTP/TCP flow lasts 230 seconds and each run of the simulation uses a different configuration combination with values chosen between:

transport protocol: ..................TCP New Reno, TCP Westwood PER between W and U:..........0.1%, 0.5%, 1.0% proxy on board: ......................split enabled, split disabled traffic direction:......................from W to G, from G to W

W: wireless device in the

urban area U: HAPS/UAV (eventually

with proxy on board) S: GEO satellite G: gateway on the ground

Fig. 5.1.1 - Simulated configuration

Each configuration has been simulated twenty times changing the seed value of the random generator. The averaged outcomes give us the packets sent in 230 seconds,

22

from this information we can provide the throughput achieved in Fig. 5.1.2 and the time required for a 5MB file transmission in Fig. 5.1.3.

Figure 5.1.2 - Avg throughput over a 230 sec transmission

Fig. 5.1.3 - Time to transmit a 5MB file from node W to G

The combination of TCP Westwood with a proxy in the HAPS/UAV produces the

highest average throughput, with values that go from 84.14% of the available bandwidth (with a PER of 1%) to 86.1% (with a PER of 0.1%). The splitting mechanism effectively hides the very high error rate present on the shortest wireless link. The average throughput achieved, in fact, remains almost constant through the various PER tested, not being affected by the transport protocol employed. The replacement of traditional TCP New Reno with TCP Westwood offers a significant advantage even if we consider the transmission time for a file. In our configuration, in fact, the latter transport protocol requires only from 65.97% (with 0.1% PER) to 46.13% (with 1% PER) of the average time needed by the former one. It is also evident that the adoption of a proxy in the HAPS/UAV that splits the connection provides an advantage, allowing TCP New Reno to achieve the same, or even slightly better, performance than does an end-to-end TCP Westwood. However, if we compare the two transmission protocols in the case in which both use the proxy in the HAPS/UAV, again the employment of TCP Westwood results in a sensible reduction of the transmission time. Compared to TCP New Reno, in fact, TCP Westwood necessitates of less time to complete the download, from 77.50% (with 0.001 PER) to 73.08% (with 0.005 PER) of the time required by the traditional transport protocol.

Figure 5.1.4 - Avg throughput over a 230 sec transmission

Fig. 5.1.5 - Time to transmit a 5MB file

from node W to G As it is easy to foresee, the good performance achieved are preserved even if we

consider a data flow traversing the links in the opposite direction: from G to W. Even in this case, TCP Westwood in maintains a correct sending rate on the link the

23

splitting scheme effectively protects the whole connection from the very high error rates present on the edge between W and U. For the four combination of transport protocol implemented and the presence or absence of the on-board proxy in node U, we present in Fig. 5.1.4 the average throughput attained on 230 seconds of simulation and in Fig. 5.1.5 the times required to transmit a 5MB file from G to W.

5.2 – Residual Capacity Estimator efficiency In the scenario depicted in Fig. 5.2.1, we consider one or more connections sharing

a bottleneck link. The accessing links to the bottleneck and the outgoing ones has very high bandwidths (100Mbps), in node A the queue is set to the pipe size while in B is set to a very large value (10,000 packets). This simplified configuration allows us to have a single point of congestion, specifically at the bottleneck link access node, and thus to properly test the ability of RCE to detect the capacity in that point. Our scheme has been tested with different values of bottleneck link capacity and presence or absence of errors. Moreover, we have also run simulations with or without a concurrent CBR (Constant Bit Rate) flow that was traversing the connection topology both in the same and in the opposite direction of the TCP flow. The transport protocol utilized is TCP Westwood with a packet size of 1000 Bytes. The CBR flow, when present, sends packets of 125 Bytes every 1ms.

Fig. 5.2.1 - The simulated environment.

We have verified the validity of the RCE outcomes running various simulations

with different configuration set up. The graphs following presented in this section show the outcomes under various conditions. In particular, the red line represents the congestion window, the green line is the slow start threshold, the yellow line shows the shared bandwidth estimated by the TCP Westwood Agile mechanism and the blue line corresponds to the capacity estimation provided by the RCE mechanism.

Fig. 5.2.2 - Single TCP flow on a bottleneck link

of 5Mbps, with no errors

Fig. 5.2.3 - Single TCP flow on a bottleneck link of 5Mbps, PER 0.1% and a period of CBR traffic

Min RTT: 70 ms

Bottleneck (various bw value) 33 ms delay

100 Mbps 1ms

100 Mbps 1ms

24

In Fig. 5.2.2 we consider a bottleneck link of 5Mbps and a single TCP flow on it, thus having the same values both for the eligible rate and for the link capacity. The RCE calculation reaches the correct value almost immediately and, specifically, 43.25 packets is the maximum value given by our estimator while the pipe size is 44 packets. We have chosen the maximum value of the RCE as a parameter in order to allow comparisons with CapProbe papers [4][5]. A more complex scenario creates Fig. 5.2.3 which shows the effectiveness of RCE, even in situation characterized by errors as those faced in a wireless environment. Indeed, it also put in evidence the prompt reactivity of our scheme when a concurrent uniformly distributed flow compares or disappears. In particular, on a channel with a PER (Packet Error Rate) of 0.1%, a single TCP flow operates from second 0 to second 30 while a CBR flow starts at second 8 of the simulation and ends at second 18. The capacity of the channel and the introduction of a uniformly distributed traffic are perfectly detected by the estimator. Our mechanism perceives the CBR flow as a decrease of the actual capacity and its behavior is not affected at all by the error losses.

Fig. 5.2.4 - 1st over 3 TCP flow on a bottleneck

link of 10Mbps, no errors, with reverse CBR flow Fig. 5.2.5 - 2nd over 3 TCP flow on a bottleneck

link of 10Mbps, no errors, with reverse CBR flow

Fig. 5.2.6 - 3rd over 3 TCP flow on a bottleneck

link of 10Mbps, no errors, with reverse CBR flow In Figg. 5.2.4, 5.2.5 and 5.2.6 we have simulated a scenario with three TCP flows

that compete for a common bottleneck of 10Mb over a total period of time of 80 seconds. For the whole period of the simulation, a CBR stream is present on the same channel but travelling in the opposite direction than the TCP flows. The TCP sources start and end to send packets at different times: the first one starts at second 0 and ends at 40, the second one transmits from second 10 to 50 and the third one starts at 20 and finish at 80. As it is easy to see, the reverse CBR traffic does not impact on the capacity estimation which is, for all the three TCP flow, promptly and correctly

25

calculated. More in detail, the maximum capacity estimated for the three connections is respectively 87.87, 86.71 and 86.84 which is very close to the factual pipe size of 88 packets.

Finally, we have also tested RCE in a scenario where traditional packet pair

techniques fail. Delayed acks employment, in fact, impacts on the acknowledgments distribution, thus affecting the detected capacity of the connection. Despite of this, when we have used delayed acks for our connections results have not been far from the correct values. In a scenario with two TCP competing for a 10Mb bottleneck over 30 seconds of simulation, the maximum capacity estimation achieved by RCE has been respectively 83.17 and 82.22 packets over a truthful capacity of 88.

27

6 – Conclusions and future works

In this work we have considered the scenario involved in the 4G development; we have presented the current state of the art, the future goals and some work we have already done toward this direction. The next generation in mobile communication will not rely on a new radio infrastructure, instead, its peculiarity will be the ability to take the best from every connection technology switching among them depending on the best options available to the user. Applications currently available only on PCs will be integrated even in smaller handheld multipurpose devices. At the same time, the Always-On and Always Best Connected philosophies behind the 4G concept, and the plethora of new services that they make possible, will constitute the real new “killer application”. On the other hand the factual establishment of this seamless ubiquitous connectivity, raises several open problems.

Up to now, our case of study has mainly been focused on the transport layer

performance in the depicted scenario. In particular in section 4 we have presented a proxy-involving architecture for providing connectivity after an urban disaster and then a new capacity estimator of the bottleneck link along the path. In the former, the combined use of HAPS/UAVs and satellites is identified as the most efficient architecture to rapidly restore service on long range connections. In this context we have compared the traditional TCP New Reno to the new TCP Westwood on a satellite link in two cases: maintaining end-to-end connection between source and destination and employing a proxy boarded on the HAPS/UAV, thus splitting the connection between the urban ground and the satellite into two parts. Results in section 5 show that the combined use of TCP Westwood and the splitting scheme attains the best performance. Not only, even singularly applied, these two techniques provide advantage against the conventional TCP New Reno on an end-to-end satellite link. Since the very good premises of this work in simulations, future augmentation could involve the use of multiple HAPS/UAVs for ground-to-ground communication also involving packets forwarding between HAPS/UAVs or, obviously, the deployment of real experiments.

The second technique presented in section 4, the capacity estimator RCE, considers

the ratio between the amount of data delivered and the effective time utilization of the channel in a RTT. Results demonstrate a quick and precise computation of the bottleneck capacity deducted the portion of the channel occupied by uniformly distributed traffic. RCE scheme is perfectly embedded in the TCP normal functioning, is not affected by random errors and only marginally by reverse data flow. The provided capacity estimation value could be used, for instance, to discriminate bad samples in TCP Westwood employment, thus improving the bandwidth estimation quality. Having a means to compute and smartly utilize the time wasted in queue by packets, RCE functionalities could be further enhanced to provide also a shared bandwidth estimation for wired-cum-wireless environments without the employment of the computationally heavy filters adopted by TCP Westwood. Future directions in this work also regard the evaluation with real experiments of the proposed scheme and a possible alternative employment for optimal data rate computation at application layer in case of video/audio streaming.

28

All the challenges listed in section 2 can be the starting point for researches that will finally converge in a comprehensive study of the 4G scenario; we have seen so far only solutions for two of them, but possible approaches can be outlined for each of those issues. Our work in the next two years will selectively cover some of the following ideas in the aim of creating factual 4G services.

For ubiquitous connectivity, universal seamless handoffs, both vertical and horizontal, are desired. An uninterrupted service development relies on the ability to switch from one access point/technology to another in a very small amount of time. Additionally, the deployment of the handoff solution should minimize the changes of the current Internet infrastructure. To achieve these goals, we could take advantages of the existing NAT [96] and IP Tunnelling [97] techniques following a scalable middleware design criteria. A proposal for a handoff architecture have to distribute the switching responsibility among three subjects: mobile host, middleware and Internet servers. The former, in fact, could autonomously determine that another access technology is more appropriate for user’s purposes and decide to switch to the new target network interface. This feature requires the ability of taking into account the diverse nature of the underlying technologies and the development of handoff timing algorithms optimized to consider a multitude of parameters as network congestion, link characteristics, energy consumption and current battery status, connection cost, user’s preferences, etc. The middleware will be in charge of adapting on-the-fly transmitted data to fit the new network conditions after the handoff. Finally, Internet servers aware of the handoff event could adapt their services to the new network condition, for instance, changing the packet emission rate, or data compression level, or information redundancy, etc.

Adaptive middleware or Internet servers could be effectively exploited in a wider

range of situations than merely during handoffs. Even without a handoff event occurrence, in fact, channel condition in a wireless environment can sensibly vary in a very short time thus suggesting a different compression/redundancy level of the information carried by the data flow. The same can be said about the earlier described situation where a user decide to continue an already running application, as video/audio streaming, on another device having significant hardware capability discrepancies.

In a wireless scenario with high mobility, vertical handoffs and on-going session

transfers between diverse devices, guaranteeing QoS raises new complex issues. First of all, a previously negotiated level of service could be no more feasible on a different access technology or due to the currently chosen device capabilities. Moreover a standard definition of Traffic Categories (TC) has not been assigned. Therefore, we have to design an edge-oriented QoS architecture able to maintain soft QoS status coherently with the media related constraints. In this configuration the proper TC can be assigned mapping the upper layer QoS parameter. Finally Content Based Policies (CBP) could be designed to provide differentiated level of importance to packets within the same flow of data. For instance, in the case of MPEG video, I-frames have to be considered much more relevant than B-frames with respect of the final perceived quality [98][99][102]. We could therefore design a scheme that puts more efforts in delivering the more important packets within a flow, for example providing retransmission only for those ones, thus reserving to this purpose a portion of the available channel. Specifically, we propose the introduction of a classification policy

29

point that uses upper layers information present in the packets to perform a flow classification at layer 2.

Platforms have to be able to seamlessly transfer their active sessions between them

every time the user decide it. We plan to implement a fast session migration scheme able to avoid loss of consistency. We believe that this feature constitutes a natural extension to the already consolidated work proposed in [82]. Not only, wireless security and an efficient access discovery/selection mechanism represent desirable integrations too. In fact, a commercial service delivery cannot leave out of consideration customers and providers’ authentication. Data integrity must be preserved in every moment, authentication and authorization protocols have to be designed to guarantee correct transactions even in an open access media as the wireless environment. For the most efficient use of the access options available the mobile terminal has to be enhanced with an optimized discovery function. This requires the design of a scheme able to map features as link qualities, pricing, operators, QoS requirements and user’s preferences on some general parameters and select the best tradeoff choice. Moreover, an easy-to-use tool could be created to support user’s intervention or initial preference settings.

Finally, GRID and P2P technologies regarding mobile terminals embody an open

field that really deserves to be studied. The former, for its attitude in providing high computational level services even to performance limited handheld devices thus exponentially augmenting their capabilities. The latter, for the natural commercial trend in offering portable multimedia devices to the market. This devices will soon completely substitute home PCs not only in the role of multimedia players but also as files downloaders. The wired consolidated scheme of these technologies has to be analysed in a wireless unstable and unpredictable channel. In particular, mobility have to be addressed to allow efficient resource discovery.

The integration and interoperability of the 4G targeted solutions that we are going

to develop, along with truly user-oriented intelligent services, will lead us to the ubiquitous Always Best Connected world that will forever enhance every aspect of our everyday life.

31

7 – Bibliography [1] http://www.isi.edu/nsnam/ns/ [2] http://www.icsi.berkeley.edu/~widmer/mnav/ns-extension/ [3] S. Bajaj, L. Breslau, D. Estrin, K. Fall, S. Floyd, P. Haldar, M. Handley, A.

Helmy, J. Heidemann, P. Huang, S. Kumar, S. McCanne, R. Rejae, P. Sharma, S. Shenker, K. Varadhan, H. Yu, Y. Xu, and D. Zappala. “Improving Simulation for Network Research”. Technical Report 99-702, University of Souther California, Mar 1999.

[4] Rohit Kapoor, Ling-Jyh Chen, M. Y. Sanadidi, Mario Gerla, “CapProbe: A

Simple and Accurate Technique to Measure Path Capacity,” Technical Report TR040001, UCLA CSD, 2004.

[5] C. E. Palazzi, C. Roseti, M. Luglio, M. Gerla, M. Y. Sanadidi, and J. Stepanek,

“Satellite coverage in urban areas using Unmanned Airborne Vehicles (UAVs)”, to appear in Proceedings of VTC2004 Spring, Milan, Italy, May 2004.

[6] T. C. Tozer and D. Grace, “High-altitude platforms for wireless

communications”, Electronics & Communication Engineering Journal, pp. 127-137, Jun 2001.

[7] D. Avagnina, F. Dovis, A. Ghiglione, P. Mulassano, “Wireless networks based

on high-altitude platforms for the provision of integrated navigation/communication services”, IEEE Communications Magazine, Volume: 40 Issue: 2, pp. 119-125, Feb 2002.

[8] Lutz, D. Cygan, M. Dippold, F. Dolainsky, W. Papke, "The Land Mobile

Satellite Channel - Recording, Statistics and Channel Model", IEEE Trans. on Vehic. Techn., Vol. VT - 40, May 1991.

[9] J. Border, M. Kojo, J. Griner, G. Montenegro and Z. Shelby, “Performance

Enhancing Proxies intended to mitigate link-related degradations”, RFC 3135, Jun 2001.

[10] J. Stepanek, Ashu Razdan, Alok Nandan, M. Gerla, M. Luglio, “The Use of a

Proxy on Board the Satellite to Improve TCP Performance”, IEEE Global Telecommunications Conference, 2002, GLOBECOM '02, Vol. 3, pp. 2950–2954, Nov 2002.

[11] S. Banerjee, B. Bhattacharjee, and C. Kommareddy, “Scalable Application

Layer Multicast”, in Proceedings of ACM Sigcomm 2002, Pittsburgh, Pennsylvania, Aug 2002.

[12] W. R. Stevens, TCP/IP Illustrated, Volume 1: The Protocols, Addison Wesley, 1994.

32

[13] E. Gustafsson and A. Jonsson, “Always Best Connected”, IEEE Wireless Communications, pp. 49-55, Feb 2003.

[14] G. Huston, “The future for TCP”, The Internet Protocol Journal, vol. 3, n. 3, Sep 2000.

[15] G. Huston, “TCP in a Wireless World”, IEEE Internet Computing, pp. 82 – 84,

Mar-Apr 2001. [16] “IEEE standard for Wireless LAN Medium Access Control (MAC) and Physical

Layer (PHY) specifications”, ISO/IEC 8802-11:1999(E), Aug 1999. [17] H. Balakrishnan, V. N. Padmanabhan, S. Sehan, and R. H. Katz, "A comparison

of mechanism for improving TCP performance over wireless links", IEEE/ACM Trans. Networking, vol. 5, no. 6, pp. 756 – 769, Dec 1997.

[18] X. Zhao, C. Castelluccia, and Mary Baker, “Flexible network support for mobile

hosts”, in MONET Special Issue on Management of Mobility in Distributed Systems, vol. 6, no. 2, 2001.

[19] S. Biaz, N. H. Vaidya, “Discriminating Congestion Losses from Wireless Losses

using Inter-Arrival Times at the Receiver”, Proc. IEEE Symposium ASSET ’99, Dallas, Texas, USA, Mar 1999.

[20] T. Goff, J. Moronski, D. Phatak, V. Gupta, “Freeze-TCP: A true end-to-end

Enhancement Mechanism for Mobile Environments”, IEEE INFOCOM 2000, Mar 2000.

[21] V. Tsaoussidis, H. Badr, “TCP-Probing: Towards an Error Control Schema with

Energy and Throughput Performance Gains”, The 8th IEEE Conference on Network Protocols, Nov 2000.

[22] S. Mascolo, C. Casetti, M. Gerla, M. Y. Sanadidi, R. Wang, “TCP Westwood:

Bandwidth Estimation for Enhanced Transport over Wireless Links”, MOBICOM 2001, Jul 2001.

[23] R. Wang, M. Valla, M.Y. Sanadidi, and M. Gerla, “Adaptive Bandwidth Share

Estimation in TCP Westwood”, Globecom 2002. [24] R. Wang, G. Pau, K. Yamada, M. Y. Sanadidi, M. Gerla, “TCP Startup

Performance in Large Bandwidth Delay Netwroks”, to appear in INFOCOM 2004, March 2004.

[25] P. Sinha, N. Venkitaraman, R. Sivakumar, V. Bharghavan, “WTCP: A Reliable

Transport Protocol for Wireless Wide-Area Networks”, ACM Mobicom '99, Aug 1999.

[26] K. Fall, S, Floyd, “Simulation-based Comparisons of Tahoe, Reno, and Sack

TCP”, ACM Computer Communication Review, Jul 1996.

33

[27] L. M. Feeney, “A QoS Aware Power Save Protocol for Wireless Ad Hoc Networks”, in Med-Hoc-Net 2002, Sep 2002.

[28] E.S. Jung and N. Vaidya, “A power control mac protocol for ad-hoc networks”,

in ACM, Mobicom, Sep 2002. [29] H. Woesner, J. P. Ebert, M. Schlaeger, and A. Wolisz, “Power-saving

mechanisms in emerging standards for wireless lans: The mac level perspective”, in IEEE Personal Communication Systems, vol.5, Jun 1998.

[30] T. Simunic, L. Benini, P. W. Glynn, and G. D. Micheli, “Dynamic power

management for portable systems”, in ACM, Mobicom, Aug 2000. [31] R. Krashinsky and H. Balakrishnan, “Minimizing energy for wireless web

access using bounded slowdown”, in ACM, Mobicom, Sep 2002. [32] L. M. Feeney and M. Nilsson, “Investigating the energy consumption of a

wireless network interface in an ad hoc networking environment”. in IEEE Infocom, Apr 2001.

[33] N. Golmie, R. E. Van Dyck, and A. Soltanian, “Interference of Bluetooth and

IEEE 802.11: simulation modeling and performance evaluation”, in ACM Int. Workshop on Modeling, Analysis, and Simulation of Wireless and Mobile Systems, Jul 2001.

[34] S. Shellhammer, “Packet Error Rate of an IEEE 802.11 WLAN in the presence

of Bluetooth”, in IEEE 802.15 Working Group Contribution, May 2000. [35] A. Kamerman, “Coexistence between Bluetooth and IEEE 802.11 CCK:

Solutions to Avoid Mutual Interference”, in IEEE 802.15 Working Group Contribution, Jul 2000.

[36] Bluetooth specifications. URL:http://www.bluetooth.com/dev/specifications.asp. [37] P. Mehta and S. Udani. “Voice over IP”, in IEEE Potentials, vol. 20, no. 4, Oct

2001. [38] N. Laoutaris and I. Stavrakakis, “Intrastream synchronization for continuous

media streams: A survey of playout schedulers”. in IEEE Network Magazine, vol. 16, no. 3, May 2002.

[39] J. Feigin and K. Pahlavan, “Measurement of Characteristics of Voice over IP in

a Wireless LAN Environment”, in IEEE Mobile Multimedia Communications (MoMuC '99), Nov 1999.

[40] T. Kanter and C. Olrog, “VoIP in Applications for Wireless Access”, in IEEE

Workshop on Local and Metropolitan Area Networks, Nov 1999. [41] K. Svanbro, J. Wiorek, and B. Olin, “Voice-over-IP-over-wireless”, in IEEE

PIMRC '00, Sep 2000.

34

[42] J. Feigin, K. Pahlavan, and M. Ylianttila, “Hardware-fitted modeling and simulation of VoIP over a Wireless LAN”, in IEEE VTC '00, Sep 2000.

[43] G. A. Mills-Tettey and D. Kotz, “Mobile Voice over IP (MVoIP): an

application-level protocol for call”, in IEEE International Conference on Performance, Computing, and Communications, Apr 2002.

[44] S. Tekinay and B. Jabbari, “Handover and channel assignment in mobile cellular

networks”, in IEEE Commun. Mag., Vol 29, No 11, pp 42-46, Nov 1991. [45] A. S. Acampora and M. Naghshineh, “An Architecture and Methodology for

Mobile-Executed Handoff in Cellular ATM Networks”, in IEEE JSAC, vol. 12, pp. 1365-75, Oct. 1994.

[46] R. Ghai and S. Singh. “An architecture and communication protocol for

picocellular networks”, in IEEE Personal Communications Magazine, pp. 36-46. [47] C. Perkins. RFC 2002 for IP Mobility Support, Oct 1996. [48] M. Stemm and R. H. Katz. “Vertical handoffs in wireless overlay networks”, in

Mobile Networks and Applications, vol. 3, no. 4, 1998. [49] M. Stemm and R. H. Katz. “Measuring and reducing energy consumption of

network interfaces in hand-held devices”, vol. E80-B, pp. 1125-31, 1997. [50] R. Mohan, J. R. Smith and C. S. Li. “Adapting multimedia internet content for

universal access”, in IEEE Transactions on Multimedia, vol. 1, no.1, 1999. [51] E. Amir, S. McCanne, and H. Zhang, “An application level video gateway”, in

Proc. ACM Multimedia '95, San Francisco, CA, USA, Nov 1995. [52] H. Bharadvaj, A. Joshi, and S. Auephanwiriyakul, “An active transcoding proxy

to support mobile web access”, in Proceedings of the 17th IEEE Symposium on Reliable Distributed Systems, 1998.

[53] R. H. Katz, “Adaptation and mobility in wireless information systems”, in IEEE

Personal Communications, Vol. 1, no. 1, 1994. [54] T. F. Abdelzaher and N. Bhatti, “Web content adaptation to improve server

overload behavior”, in Proceedings of the 8th International WWW Conference, Toronto, Canada, May 1999.

[55] A. Fox, S. Gribble, Y. Chawathe and E. Brewer “Adapting to network and client

variation using active proxies: Lessons and perspectives”, In Special issue of IEEE Personal Communications on Adaptation, Aug 1998.

[56] X. Zhao, C. Castelluccia, and M. Baker, “Flexible network support for mobile

hosts”, in MONET Special Issue on Management of Mobility in Distributed Systems, vol. 6, no. 2, 2001.

35

[57] R. Chakravorty and I. Pratt. “WWW Performance over GPRS”, in IEEE MWCN '02, Stockholm, Sweden, Sep 2002.

[58] B. Kim Z. Jiang, L. Chang and K. Leung, “Incorporating Proxy Services into

Wide Area Cellular IP Networks”, in IEEE WCNC '00, Sep 2000. [59] Z. Jiang and K. Leung, “Link Condition Based Proxies for QoS Management in

Wireless Networks”, in PIMRC '01, Sep 2001. [60] S. Gitzenis and N. Bambos, “Power-Controlled Data Prefetching/Caching in

Wireless Packet Networks”, in INFOCOM '02, Jun 2002. [61] M. Reisslein and K. Ross, “High-Performance Prefetching Protocols for VBR

Prerecorded Video”, in IEEE Network, vol. 12, no. 6, Nov/Dec 1998. [62] F. H. P. Fitzek and M. Reisslein, “A Prefetching Protocol for Continuous Media

Streaming in Wireless Environments”, in IEEE Journal on Selected Areas in Communications, vol. 19, no. 10, 2001.

[64] K. Brown, S. Singh, “M-TCP: TCP for Mobile Cellular Networks”, ACM

Computer Communication Review, Vol. 27, n. 5, Jul 1997. [65] C. Dovrolis, P. Ramanathan, and D. Moore, “What do packet dispersion

techniques measure?", in Proceedings of IEEE Infocom'01, Anchorage, Alaska, Apr 2001.

[66] M. Stemm and R. H. Katz, "Vertical Handoffs in Wireless Overlay Networks",

ACM Mobile Networks and Applications (MONET) Journal, 1998, Special Issue on Mobile Networking in the Internet.

[67] IEEE Computer Society LAN MAN Standards Committee, “Wireless LAN

Medium Access Control (MAC) and Physical Layer (PHY) Specifications”, the Institute of Electrical and Electronics Engineers, IEEE Std 802.11, Jun 1997.

[68] T. Phan, G. Zorpas and R. Bagrodia, "The Convergence of Heterogeneous

Internet-Connected Clients Within iMASH", IEEE Personal Communications magazine, Apr 2002.

[69] The iMASH Project, URL: http://pcl.cs.ucla.edu/projects/imash/index.htm [70] The MosquitoNet Project, URL: http://mosquitonet.stanford.edu/index.html [71] C. Boyd and A. Mathuria, “Key establishment protocols for secure mobile

communications: a critical survey”, Computer Communications, Vol. 23, pp. 575-587, 2000.

[72] K. Wang, S.K. Tripathi, "Mobile-End Transport Protocols: An Alternative to

TCP/IP Over Wireless Links", in Proc. IEEE Infocom ‘98, San Francisco, California, USA, pp. 1046-1053, Apr 1998.

36

[73] W. K. Wong, H. Zhu, V. C. M. Leung, “Soft QoS Provisioning Using he Token Bank Fair Queuing Scheduling Algorithm”, IEEE Wireless Communication magazine, Jun 2003.

[74] W. Pattara-Atikom, P. Krishnamurthy and S. Banerjee, “Distributed Mechanisms for Quality of Service in Wireless LANs”, IEEE Wireless Communication magazine, Jun 2003.

[75] P. Mohapatra, J. Li and C. Gui, “QoS in Mobile Ad Hoc Networks”, IEEE

Wireless Communications magazine, Jun 2003. [76] R. Bagrodia, T. Phan and R. Guy “A Scalable, Distributed Middleware Service

Architecture to Support Mobile Internet Applications”, Wireless Networks, the Journal of Mobile Communication, Computation and Information (WINET), Jul 2003.

[77] K. Ohta, T. Yoshikawa, T. Nakagawa, Y. Isoda and S. Kurakake, “Adaptive

Terminal Middleware for Session Mobility”, 23rd International Conference on Distributed Computing Systems Workshops (ICDCSW’03), Providence, Rhode Island, USA, May 2003.

[78] E. Skow, J. Kong, T. Phan, F. Cheng, R. Guy, R. Bagrodia, M. Gerla and S. Lu,

“A Security Architecture for Application Session Handoff”, International Conference on Communications (ICC 2002), Apr 28 – May 2, 2002.

[79] A. M. Basyouni and S. E. Tavares, “Public Key versus Private Key in Wireless

Authentication Protocols”, in Proceedings of the Canadian Workshop on Information Theory, Toronto, Canada, Jun 1997.

[80] S.H. Park, A. Ganz and Z. Ganz, “Token-Based Security Protocol for Wireless

Local Area Networks”, 3rd Telecommunications R&D Conference in Massachusetts , Lowell, MA, Nov 1997.

[81] J. Jun and M. L. Sichitiu, “The Nominal Capacity of Wireless Mesh Networks”,

IEEE Wireless Communications magazine, Oct 2003. [82] V. Ghini, G. Pau, M. Roccetti, P. Salomoni, M. Gerla, “For Here or To Go?

Downloading Music on the Move with an Ultra Reliable Wireless Internet Application”, IEEE ICC’2004, Paris, 2004.

[83] I. Foster and C. Kesselman, editors. The Grid: Blueprint for a New Computing

Infrastructure. Morgan-Kaufmann, 1999. [84] K. Czajkowski, S. Fitzgerald, I. Foster and C. Kesselman, “Grid Information

Services for Distributed Resource Sharing”, in Proc. 10th IEEE Symposium on High Performance Distributed Computing, 2001.

[85] T-Mobile HotSpot, URL: http://www.t-mobile.com/hotspot/ [86] UMTS Forum, URL: http://www.umts-forum.org

37

[87] P. Sinha, N. Venkitaraman, R. Sivakumar and V. Bharghavan, “WTCP: A Reliable Transport Protocol for Wireless Wide-Area Networks”, ACM Mobicom '99, Aug 1999.

[88] Napster Home Page, http://www.napster.com/ [89] Overview of IEEE 802.11b Security, URL: http://developer.intel.com/technology/itj/q22000/articles/art_5.htm [90] I. Foster and A. Iamnitchi, “On Death, Taxes, and the Convergence of Peer-to-

Peer and Grid Computing”, in 2nd International Workshop on Peer-to-Peer Systems (IPTPS'03), Berkeley, CA, Feb 2003.

[91] S. Ratnasami, P. Francis, M. Handley, R. Karp and S. Shanker, “A Scalable

Content Addressable Network”, in Proc. of SIGCOMM’01, San Diego, CA, Aug 2001.

[92] A. Rowstron and P. Drushel, “Pastry: Scalable Distributed Object Location and

Routing for Large Peer-to-Peer Systems”, in Proc. IFIP/ACM Middleware 2001, Heidelberg, Germany, Nov 2001.

[93] I. Stoica, R. Morris, D. Karger, M. F. Kaashoek and H. Balakrishnan, “Chord: A

Scalable P2P Lookup Service for Interent Applications”, in Proc. of SIGCOMM’01, San Diego, CA, Aug 2001.

[94] I. Foster, C. Kesselman and S. Tuecke, "The Anatomy of the Grid: Enabling

Scalable Virtual Organizations", International Journal of Supercomputer Applications, vol. 15, no. 3, 2001.

[95] M. Roccetti, V. Ghini and P. Salomoni, “Distributing Music from IP Networks

to UMTS Terminals: an Experimental Study'', in Proc. 2002 SCS Euromedia Conference (EUROMEDIA2002), (M. Roccetti Ed.) The Society for Modeling and Simulation International, Modena (Italy), Apr 2002, pp. 147-154.

[96] P. Srisuresh and M. Holdrege, “IP Network Address Translator (NAT)

Terminology and Considerations”, RFC 2663, Aug 1999. [97] C. Perkins, “IP Encapsulation within IP”, RFC 2003, Oct 1996. [98] Coding of Moving Pictures and Associated Audio for Digital Storage Media at

up to About 1.5 Mbits/s, (MPEG-1), Oct 1993. [99] Generic Coding of Moving Pictures and Associated Audio, ISO/IEC 13 818-2,

(MPEG-2), Nov 1993. [100] C. E. Palazzi, “802.11 ARQ Scheme Analysis in presence of Data and Stream

Traffic”, submitted at WCC2004-Student-Forum IFIP World Computer Congress 2004, Tolosa, Aug 2004.

38

[101] C. E. Palazzi, “Residual Capacity Estimator for TCP on wired/wireless links”, submitted at WCC2004-Student-Forum IFIP World Computer Congress 2004, Tolosa, Aug 2004.

[102] Coding of Moving Pictures and Associated Audio MPEG98/W2194, (MPEG-4),

Mar 1998. [103] C.H. Nam, Soung C. Liew, C.P. Fu, “An Experimental Study of ARQ Protocol

in 802.11b Wireless LAN”, in Proc. of IEEE VTC 2003, 2003. [104] G. Xylomenos and G. C. Polyzos, “TCP and UDP Performance over a Wireless

LAN”, in Proc. of IEEE INFOCOM ’99, 1999. [105] S. Floyd, T. Henderson, “The NewReno Modification to TCP’s Fast Recovery

Algorithm”, IETF RFC 2582, April 1999. [106] A. Balk, M. Gerla, M. Sanadidi, D. Maggiorini “Adaptive Video Streaming:

Pre-encoded MPEG-4 with Bandwidth Scaling’, to appear in Computer Network Journal.

[107] M. Gerla, B. K. F. Ng, M. Y. Sanadidi, M. Valla, R. Wang “TCP Westwood

with Adaptive Bandwidth Estimation to Improve Efficiency/Friendliness Tradeoffs”, to appear in Computer Communication Journal.

[108] S. Mascolo, A. Grieco, G. Pau, M. Gerla, C. Casetti, “End-to-End Bandwidth

Estimation in TCP to Improve Wireless Link Utilization”, European Wireless Conference, Florence, Italy, Feb 2002.

[109] R. Wang, M. Valla, M.Y. Sanadidi, B. K. F. Ng, M. Gerla,

“Efficiency/Friendliness Tradeoffs in TCP Westwood”, Seventh IEEE Symposium on Computers and Communications, Taormina, Italy, Jul 2002.

[110] J. C. Haartsen, “The Bluetooth Radio System”, IEEE Personal Communications,

Vol. 7, No.1, Feb 2000, pp. 28-36. [111] M. Meyer, J. Sachs and M. Holzke, “Performance Evaluation of a TCP Proxy in

WCDMA Networks”, in IEEE Wireless Communication, Oct 2003. [112]J. Weatherall and A. Jones, “Ubiquitous Networks and their applications”, in

IEEE Wireless Communication, Feb 2002. [113]N. Davies, K. Cheverst, A. Friday and K.Mitchell, “Future Wireless

Applications for a Networked City: Services for Visitors and Residents”, Feb 2002.

[114] V. Ghini, G. Pau, M. Roccetti, P. Salomoni and M. Gerla, “Smart Download on

the Go: A Wireless Internet Application for Music Distribution over heterogeneous Networks”, in Proc. IEEE International Conference on Communications - Access and Home Networks Symposium (ICC'04), accepted for publication, Paris, France, IEEE Communications Society, Jun 2004.