An adaptive QoS framework for integrated cellular and WLAN networks

Computer Networks 47 (2005) 167–183 www.elsevier.com/locate/comnet

An adaptive QoS framework for integrated cellular and WLAN networks Xin Gang Wang a,*, Geyong Min a, John E. Mellor a, Khalid Al-Begain b, Lin Guan a a

Department of Computing, School of Informatics, University of Bradford, Bradford BD7 1DP, UK b School of Computing, University of Glamorgan, Wales CF37 1DL, UK Available online 6 August 2004

Abstract The design of a network architecture that can efficiently integrate WLAN and cellular networks is a challenging task, particularly when the objective is to make the interoperation between the two networks as seamless and as efficient as possible. To provide end-to-end quality of service (QoS) support is one of the key stages towards such a goal. Due to various constraints, such as the unbalanced capacity of the two systems, handoff from user mobility and unreliable transmission media, end-to-end QoS is difficult to guarantee. In this paper, we propose a generic reservation-based QoS model for the integrated cellular and WLAN networks. It uses an adaptation mechanism to address the above issues and to support end-to-end QoS. The validity of the proposed scheme is demonstrated via simulation experiments. The performance results reveal that this new scheme can considerably improve the system resource utilization and reduce the call blocking probability and handoff dropping probability of the integrated networks while maintaining acceptable QoS to the end users.
2004 Elsevier B.V. All rights reserved. Keywords: WLAN and cellular network integration; QoS framework; Reservation; Bandwidth adaptation

1. Introduction *

Corresponding author. E-mail addresses: [email protected] (X.G. Wang), [email protected] (G. Min), [email protected] (J.E. Mellor), [email protected] (K. Al-Begain), l.guan@ bradford.ac.uk (L. Guan).

In the future, wireless service provision will be characterised by global mobile access anywhere and anytime [1]. Two major access technologies for those mobile communication systems are wireless local area networks (WLAN) and cellular

1389-1286/$ - see front matter
2004 Elsevier B.V. All rights reserved. doi:10.1016/j.comnet.2004.07.003

168

X.G. Wang et al. / Computer Networks 47 (2005) 167–183

networks, such as global system for mobile communication (GSM), general packet radio service (GPRS) and universal mobile telecommunications system (UMTS). WLAN systems provide very high data rates at a relatively low cost compared to cellular networks and is becoming more and more popular. However, WLAN technology is more likely a complimentary access method rather than a competitor to 3G networks because it has limited coverage area and less support for high speed mobility. So far WLANs have been setup in places like airports, hotels, and campuses. 3G networks are gradually deploying worldwide. Interconnecting WLAN radio access networks and 3G cellular networks with QoS support offer an efficient way to enhance the network operator service. The communication systems dominated by voice transmission employed circuit-switching technology [2] for a long time. However, the demand for data communication is increasing and drives the development of the packet switching technology, which led to the born of the Internet. Currently, the mobile communication system is facing the same evolution as the Internet. The network operators are migrating from circuitswitched GSM systems to GPRS and 3G networks worldwide [3]. The ultimate vision is to provide a universal all-IP platform. The integration of WLAN and cellular networks is an important step for this process. It can provide end users with benefits like lower cost of transmission and higher bandwidth without losing the roaming features or pervasive aspects now emerging. However, the design of a network architecture that efficiently integrates WLAN and cellular networks is a challenging task, particularly when the objective is to make the interoperation of the two technologies as seamless and as efficient as possible [4]. To provide end-to-end quality of service (QoS) support is one of the key issues in the design of integrated WLAN and cellular networks. Two major models for QoS support have been proposed in the network research community. One is based on reservation and another is based on prioritization, namely: IntServ and DiffServ [5]. They differ in that the reservation-based approach sends

signals through the data path and books its QoS requirements before the actual data transmission, while the prioritization-based approach simply marks the traffic on an individual packet basis to indicate the QoS requirements and sends the packets to the network. It is well known that the Internet has some fundamental scalability limitations when it comes to the management of individual traffic flows using the reservation approach. Its successor, the prioritization approach, addresses the scalability problem at the cost of coarser service granularity. Many difficulties emerge when attempting to provide QoS solutions for integrated WLAN and cellular networks owing to the unbalanced capacity of the two systems, issues raised by handover between homogeneous cells and heterogeneous cells caused by user mobility, and transmission through the unreliable wireless media. To enable efficient use of the scarce resources provided by cellular networks while also maintaining strong QoS guarantees, we propose a generic reservation-based QoS model for the integrated cellular and WLAN network. Under the proposed QoS framework, we develop an adaptation mechanism to address the various challenges in the integrated mobile networks. The validity of the proposed frame-work is demonstrated through simulation experiments. The performance results indicate that this new scheme can improve the system resource utilization and considerably reduce the call blocking probability and the handover dropping probability of the integrated network while still maintaining acceptable QoS to the end users. The rest of this paper is organized as follows. Section 2 reviews the existing QoS architectures and mechanisms which are essential for the following analysis. Section 3 presents the problems related to QoS support over the integrated system. We introduce and analyze the proposed QoS framework in Section 4. An adaptive algorithm to manage QoS in the framework is introduced in Section 5. Section 6 describes the simulation and discusses the performance results based on the proposed framework. Section 7 summarizes this study and gives concluding remarks.

X.G. Wang et al. / Computer Networks 47 (2005) 167–183

2. Preliminary work 2.1. WLAN Qos The success of the Internet and the availability of inexpensive WLAN equipment has spurred the demand for mobile data access. WLANs often operate on a centralized architecture, where an access point (AP) coordinates mobile terminals (MT) in accessing the wireless medium and links the traffic into the wired network [6]. The AP is either a layer 2 bridge between IEEE 802.11 and Ethernet or a layer 3 router between IEEE 802.11 and a backbone network. The MT is typically a laptop computer or a personal digital assistant (PDA) with a built-in WLAN radio module or a WLAN card. There are two major WLAN standards: IEEE 802.11 and HiperLAN. The success of the IEEE 802.11 in the marketplace has made it the de facto WLAN standard worldwide. Even for the IEEE 802.11 standards used today, there are a few variations: the widely deployed 11 Mb/s IEEE 802.11b [6], the high speed (54 Mb/s) version 5 GHz IEEE 802.11a and its cousin IEEE 802.11g. From the WLAN system point of view, the consequences of these upgrades are mostly limited to the radio interface as higher layers remain unchanged and they only support the best effort service to the end users. To address the QoS requirements, a supplement standard IEEE 802.11e is under development [7,8]. Two new mechanisms are defined for QoS support, namely enhanced distributed coordination function (EDCF) and hybrid coordination function (HCF) [7]. EDCF is a basic QoS supporting mechanism. It can provide differential of service (DoS) [9] and it is still contention-based, while HCF works as a guaranteed method to provide

QoS. In the standard, an area covered by a 802.11g network is called a quality basic service set (QBSS), which is often composed of a hybrid coordinator (HC) and some 802.11e-compliant enhanced stations. The HC can be any station in the QBSS which can work as the central coordinator, but it typically resides within an 802.11e AP. EDCF is a contention-based medium access method and QoS support is realized with the introduction of traffic categories (TCs) and access categories (ACs). There are eight TCs to provide differentiated distributed access to the wireless medium. They are the same as defined in the IEEE 802.1d bridge standard for reasons of consistency. These eight TCs are mapped into four ACs as shown in Table 1. The access priority for different traffic classes is controlled both by a given different contention window (CW) and by a different inter frame space (IFS) to different ACs as illustrated in Fig. 1. Each AC is characterised with an arbitration inter frame space (AIFS) and a persistence factor (PF). The higher AC has a lower AIFS and a smaller PF compared to lower ACs. Their formulation is listed below: AIFSD½AC ¼ SIFS þ AIFS½AC  Slottime;

ð1Þ

newCW½ACPððoldCW½AC þ 1Þ  PFÞ  1:

ð2Þ

The HCF serves as an extension for the EDCF and it has both contention-based and controlled contention-free channel access methods in a single channel access cycle. Each transmission cycle is realized in the form of a superframe as shown in Fig. 2 which consists of a contention period (CP) and a contention free period (CFP). The EDCF

Table 1 Traffic categories and access categories Traffic categories (TCs)

Access categories (ACs)

0 1 2 3 4 5 6 7

0 0 0 1 2 2 3 3

(Default) Best effort Background Standard (spare) Excellent effort (business critical) Controlled load (streaming multimedia) Video (interactive media) Voice (interactive voice) Network control reserved traffic

169

Best effort Best effort Best effort Video probe Video Video Voice Voice

170

X.G. Wang et al. / Computer Networks 47 (2005) 167–183 AIFS[j] AIFS[i]

Immediate access when

DIFS

Medium is free >= DIFS/AIFS[i] DIFS/AIFS

Contention Window

PIFS SIFS

Busy Medium

Next Frame

Backoff Slots Slot time

Select Slot and Decrement Backoff as long as medium is idle

Defer Access

Fig. 1. Different IFSs and CWs for different ACs.

802.11E Periodic Superframe Contention Free Period (CFP)

Beacon

HC Poll

Contention Period (CP)

CFP-end

TBTT

TXOP

RTS/CTS/fragment

TXOP

DATA/ACK

Time

Fig. 2. HCF superframe structure.

operates in the CP and a controlled channel access mechanism called polling operates concurrently within the CFP. Because the CFP uses a shorter IFS called polling IFS (PIFS) than the CP, HCF always has priority over the EDCF method. During a superframe, a HC sends out a QoS CF-Poll which includes the transmission order and the maximum transmission time. A station will not access the channel unless it receives such a polling packet. This HC controlled channel access method guarantees time-bounded service to QoS applications. 2.2. UMTS QoS The UMTS is the widely accepted 3G cellular network standard and has a layered architecture for the support of end-to-end QoS for the packet data domain [10,11]. Fig. 3 shows the UMTS

QoS architecture. Each module calls its bearer service (BS) to accommodate the QoS requirements, when making a QoS transmission. A BS has a basic functionality defined in each layer featured by different parameters like traffic type, traffic characteristics and supported bit rate. It includes all aspects to enable the provision of a contracted QoS. These aspects are, among others, the control signaling, user plane transport and QoS management functionality. There are various BS managers in the different modules to coordinate the overall management procedures. A signaling protocol then can call these BS managers to accommodate the requested QoS [12]. Packet data protocol (PDP) is used to establish the QoS connection within the UMTS network. If the destination is an address outside the UMTS network, the external BS manager which resides

X.G. Wang et al. / Computer Networks 47 (2005) 167–183 TE

MT

CN Iu EDGE NODE

UTRAN

CN Gateway

171 TE

End-to-End Service

TE/MT Local Bearer Service

External Bearer Service

UMTS Bearer Service

Radio Access Bearer Service

Radio Bearer Service

UTRA FDD/TDD Service

Iu Bearer Service

CN Bearer Service

Backbone Bearer Service

Physical Bearer Service

Fig. 3. UMTS QoS architecture.

in the gateway GPRS support node (GGSN) has to be used to control IP bearer services by standard IP mechanisms. Before a terminal equipment (TE) can send out actual data traffic, it has to send a PDP request packet through the entire data path and a QoS path is established while receiving a PDP acceptance packet. There are four basic types of traffic classes defined in UMTS [13], namely conversational, streaming, interactive and background. Conversational and streaming classes are for real-time applications, while interactive and background classes are used for delay-tolerant applications. Conversational class is the most challenging class and only a very short delay and negligible delay jitter are acceptable. As the trend is to an all-IP network, this class is expected to support voice over IP for radio applications. The streaming class has fewer requirements than the conversational class although still in the real-time catalog. A larger buffer is arranged on the receiver side to remove the delay variations. These UMTS QoS classes are summarized in Table 2. 2.3. WLAN and UMTS Integration The early work on the integration of WLAN and 3G networks was done by the ETSI BRAN

project [14]. Two different fundamental methods have been proposed for merging WLAN and cellular networks namely loose coupling and tight coupling [14]. Loose coupling is shown in Fig. 4 and it features less integration between the two types of networks, as its name implies. In this scenario, the WLAN and cellular networks are two separate access networks. The WLAN access network is attached to the Internet backbone, and the cellular networks into the cellular core network. The access networks do not have anything in common, but the core networks are connected together. Without necessarily modifying the 3G core network, a loosely coupled WLAN and 3G network can use existing mechanisms to accommodate its usersÕ needs, for instance, using an authentication, authorization and accounting (AAA) server to handle the user subscription to these networks and using mobile IP (MIP) to facilitate userÕs roaming among different access networks [4]. The motivation is to try and minimize the changes to the cellular core networks, therefore reducing the cost of this solution. Tight coupling illustrated in Fig. 5 suggests that WLAN technology is employed as a new radio access technology within the cellular system. Regardless of the access technology, there would only be one common cellular core network. This can be

Background download of e-mail, electronic postcard Web browsing, network games Streaming multimedia Voice, video telephony, video games

•

Fig. 4. Loose coupling.

Application examples

Preserve data integrity • • Use of buffer to smooth out jitter •

Preserve data integrity

Destination is not expecting the data within a certain time Asymmetric applications, more tolerant to jitter than conversational class

Request response pattern

Interactive

•

Streaming

•

Preserve time relation (variation) between information entities of the stream Conversational pattern (stringent and low delay)

Conversational

• Characteristics

Traffic class

Table 2 UMTS QoS traffic classes

Background

X.G. Wang et al. / Computer Networks 47 (2005) 167–183

•

172

done by connecting a WLAN AP to a radio network controller (RNC) via the Iu bearer interface. Another possible approach is that the whole radio access network (including the base station controller) is WLAN specific and it would attach into the core network via an Iu interface. Since the core network has to be directly exposed to the WLAN for tight coupling, the same operator must own both the WLAN and 3G networks and this makes the integration of independently operated WLAN with the 3G networks not possible. 3GPP has recently also taken the initiative to develop a cellular–WLAN interworking architecture [10]. This interworking architecture is based on loose coupling and introduces the authentication, authorization and accounting (AAA) service and mobile IP (MIP) functionality into the 3GPP standards. The entire integration is achieved without setting any 3GPP-specific requirements on the

Fig. 5. Tight coupling.

X.G. Wang et al. / Computer Networks 47 (2005) 167–183

WLAN systems, but relying on the existing functionality available in a typical WLAN access network.

3. QoS issues in integrated WLAN and UMTS Supporting QoS for integrated WLAN and UMTS networks is a challenging task. In the fixed broadband, admitted resources for a QoS connection remain relatively static, since there is no user movement or any radio fading problem. Unlike in homogeneous wired networks, providing QoS for integrated WLAN and UMTS networks has some fundamental bottlenecks [15]. Firstly, WLAN and UMTS networks have different transmission capacity over the radio interfaces, therefore the handoff between the two systems makes the maintenance of QoS connection very hard. WLAN can provide a transmission speed from 11 Mb/s up to 54 Mb/s theoretically, while UMTS has only 144 kb/s at vehicular speed, 384 kb/s outdoor to indoor and pedestrian and 2 Mb/s indoor. If we keep the QoS resource assigned by UMTS to a connection when it is actually in a WLAN hotspot, the advantage of the high speed of WLAN is not fully taken. On the other hand, if we use a WLAN parameter for a station in UMTS network, the connection may not be admitted at all. Therefore, to maintain a sensible QoS framework, one has to consider the significant different transmission capacity between two systems especially when user handoff takes place. The second constraint is that WLAN operates on a free ISM band and has a lot of uncontrollable interference (i.e., microwave), although some techniques like spread spectrum are used to reduce the interference. Such kinds of problems are beyond engineering control and a hard QoS guarantee is very difficult or even impossible to achieve under such circumstances. To support QoS in packet switching networks, there has to be some mechanism to control network load under a threshold so that the system can achieve a satisfied performance. The third bottleneck is that the achievable QoS levels in WLAN and 3G cellular networks do not match each other.

173

3G cellular networks are very well designed with careful network planning and mature admission control algorithms. Therefore, the achievable QoS level is relatively high, while 802.11e WLAN works under a more robust environment and it is difficult to achieve hard QoS, although some form of admission control [7] has been provided for HCF in the IEEE 802.11e standard. Even the EDCF can only provide differential of service (DoS). All these problems lead us to find an adaptive solution for integrated WLAN and cellular networks, which can address the above issues and also provide practical and user-satisfying QoS.

4. An adaptive QoS architecture Increasing data service requirements and Internet applications are driving the cellular network evolving into an IP-based packet switching network [3]. Our proposed QoS framework is based on a packet switching core network with the UMTS architecture. However, it holds as well with the GPRS 2.5G networks or other packet switching cellular systems. The overall architecture is shown in Fig. 6. It is well known that the Internet has some fundamental scalability limitations [5] when it comes to manage individual traffic flows using the reservation-based approach. Its successor, the prioritization approach addresses the scalability problem at the cost of coarser service granularity. To enable efficient use of scarce resources provided by the cellular networks while also maintaining strong service guarantees, we adopt the reservation-based approach [16]. In WLAN the reservation is achieved by using the HCF and in UMTS is achieved by the functionality provide by BS. The other components of the framework are defined below: • A policy provisioning module (PPM) The PPM is responsible for mapping actual user QoS profiles with their subscription information and decides the traffic classes for the user traffic flows. Then these QoS parameters can be handed to the connection admission control module (CAC) to process.

174

X.G. Wang et al. / Computer Networks 47 (2005) 167–183

Fig. 6. Proposed QoS architecture.

• A connection admission control module (CAC) The CAC is to admit the number of flows that can be served and allocates bandwidth to them through signaling to all the network nodes along the traffic path. It also needs to maintain the QoS requirements of existing connections. • A QoS mobility management module (MMM) The MMM decides whether terminals are detached, connected or idle and also monitors active nodes moving at high speed. • A QoS monitoring module (monitor) The monitor continuously measures whether the QoS requirements of mobile nodes have been satisfied. The components illustrated are viewed as logical entities. These components can be actually implemented combined with realistic network components or in an independent location. A combined IntServ and DiffServ method is adopted when connecting the integrated network to the Internet backbone in order to address the scalability problem.

4.1. Components analysis 4.1.1. QoS policy provisioning UsersÕ context with the QoS requirement is first issued to PPM where the usersÕ subscribed information together with traffic classes is examined. Then a QoS signal with suggested degradation profile is made and sent to both the end user and CAC module (Fig. 7).

Fig. 7. A PPM structure.

X.G. Wang et al. / Computer Networks 47 (2005) 167–183

• Degradation profile The negotiation of established QoS connection is allowed through the degradation profile. When the user requests to establish a QoS call, certain network resources need to be admitted. The requested QoS has to be allocated when the connection is setup. If certain conditions change over the activation time, a negotiation procedure is called. The degrade profile can include the following QoS attributes: – the minimum acceptable rate (bit/s), – the bit error rate (BER) or frame error rate (FER), – the maximum loss ratio (the proportion of received packets to undelivered packets), – the maximum tolerated delay (ms), – the maximum tolerated jitter (ms) (the variation in delay). • UsersÕ subscription information Practical service solutions should provide flexible ways of delivering QoS to the users. For example, the network operator could offer four packages: gold, silver, bronze and pay as you go. The gold package would allow users to transmit at the maximum rate 5 mb/s in the hot-spot area plus other calling features. Silver is maximum 1 mb/s and bronze is just best effort service. This information can be accessed by the PPM to identify and mark individual traffic flows for coordinating QoS from end to end between network elements.

way until reaching the destination. This reservation information can either be hard or soft in the router buffer. Hard state means that the state information stored in the router has to be removed by an explicit signaling message while soft state has a timeout field and removes itself when this value gets to zero. The well known scalability (or known as state explosion) problem with the reservation approach, limits the domain of this solution to small networks. However, soft state can be used to effectively increase the network scalability. On the other hand, hard state cannot only reduce the amount of signaling but also guarantee userÕs QoS profiles. These trade-offs should be considered together with the practical factors of some particular networks. • Degradation of other connections Applying QoS means treating some traffic in preference to others, and this implies the ability to reject traffic. Especially in wireless mobile communication networks, uncontrolled error rate and usersÕ mobility make us have to look for adaptation solutions. The use of the degradation profile provides us a gradation between different QoS merits; negotiation between different network flows is an effective way to improve the overall system performance. • QoS structure within a single network node QoS within a single network element is illustrated in Fig. 8. When there is a new connection or a handoff connection, the request is submitted to the CAC and then the CAC invokes the signaling protocol RSVP to book the required

4.1.2. QoS connection admission control (CAC) The CAC module receives a connection request from the PPM along with the QoS requirements. It consults with the MMM to get the status of user mobility. Then CAC uses some reservation protocols, RSVP, for example, to book the actual resource for usersÕ flow. Based on RSVP signaling feedback, the connection is finally granted, declined or renegotiated. Some related issues are discussed below: • Required resources available RSVP is an end to end signaling protocol. It reserves necessary network resources along its

175

CAC Signalling Routing network RSVP

Data In

Forwarding Table

Perflow QoS

Packet Forwarding

Packet Scheduling

Fig. 8. QoS structure inside a network node.

Data Out

176

X.G. Wang et al. / Computer Networks 47 (2005) 167–183

QoS resource along the whole data path. An adaptation mechanism, which integrates the user degradation profile into its QoS profile, can be used along with the RSVP signaling protocol to improve the probability of successful operations. Finally, the result of the RSVP signaling is returned to CAC and a decision is made whether to accept this connection or reject it.

4.1.3. QoS mobility management module (MMM) UsersÕ mobility has a significant impact on the QoS of CAC and it plays an important role in the model. Within an integrated cellular and WLAN network, a handoff can occur when a mobile node enters a hot-spot area or when it decides to leave the hot-spot area. Because hot-spots are usually within the coverage of cellular networks, the actual handoff is not necessary and a decision should be made based on user desire. We name this user-triggered handoff or desirable handoff (DH). Note it is different from general term of handoff, because it is not time critical as the mobile nodes can be connected to WLAN and cellular networks simultaneously (Fig. 9). • Roaming from cellular networks into WLAN A DH may occur when a mobile node roams into a WLAN. This implies that the route taken by data will change. Any QoS that was established for that data flow will be disrupted. A simple solution to this problem is to establish a new WLAN reservation before handing the mobile node over to the WLAN, because DHÕs time tolerance makes this approach realistic. As the wireless link bandwidth will rise dramatically, the new submitted QoS profile should consider usersÕ subscription status and give an appropriate request.

Fig. 10. Normal handoff.

• Roaming from WLAN into cellular networks A normal handoff occurs when a mobile node roams from WLAN into a cellular network (Fig. 10). A new reservation has to be made again. Moreover the actual handoff time needs to be kept tightly in order to provide seamless service. Since the network resources reserved by the user in the WLAN is normally over the UMTS capacity, the actual handoff dropping probability could be very high. An adaptation mechanism needs to be embedded in CAC and this module can reduce the QoS request by using the degradation profile. Therefore, the system performance can be improved without losing acceptable QoS level. • Speed Considering the limited coverage area of WLAN, a user moving at high speed could experience handoff too frequently to register with a WLAN system. Therefore, some kind of speed measurement could be defined in MMM. A threshold value could also be determined to prevent such an undesirable handoff from occurring.

4.1.4. QoS monitoring Once the streaming data is flowing, traffic meters measure its temporal properties against the QoS contract. If the QoS profile established by end users is not satisfied, this monitor may pass state information to CAC or other components to trigger specific actions. This feedback approach enables the QoS to adapt to the dynamic changes in the networks. 4.2. Connection with the IP backbone

Fig. 9. Desirable handoff.

This section describes the QoS architecture when the integrated networks are connected to

X.G. Wang et al. / Computer Networks 47 (2005) 167–183

177

End to End RSVP signaling

IntServ Domain

DiffServ Core

IntServ Domain

Fig. 11. Combined IntServ and DiffServ architecture.

an IP backbone. The main focus of this network QoS mechanism is to provide IntServ flexible QoS definition while not losing scalability. The overall architecture is shown in Fig. 11. Mobile terminals (BT) and local network operators implement IntServ and core network operators implement the DiffServ architecture. RSVP is used as the signaling protocol for real-time services. Signaling takes place between end nodes and IntServ edge routers. Some IntServ routers also act as DiffServ edge routers. From the perspective of an IntServ network, these routers simply tunnel through a non-IntServ region. From the perspective of DiffServ routers, the IntServ network is not visible and treated as normal traffic with priorities. 4.3. QoS class mapping To provide unified QoS traffic classes, the QoS traffic classes from UMTS and WLAN are mapped into a new set of QoS traffic classes namely: broadband conversational (B-conversational), broadband streaming (B-streaming), narrowband conversational (N-conversational), narrowband streaming (N-streaming), interactive and background. The mapping relationships are shown in Table 3. Table 3 QoS classes mapping table Class

Integrated network

WLAN

UMTS

1 2 3 4 5 6

B-conversational B-streaming N-conversational N-streaming Interactive Background

Voice Video – Video probe – Best effort

– – Conversational Streaming Interactive Background

5. The adaptation algorithm The bandwidth adaptation algorithm, as the key factor of the proposed framework, decides how to adjust the QoS connections since mobile users should be able to seamlessly maintain their ongoing sessions at a satisfactory level. Ideally each call in the system should be allocated the maximum allowable bandwidth. However, WLAN and cellular networks have different transmission capacities; a session that consumes a moderate amount of bandwidth in a WLAN system can be greedy and therefore could be rejected in the cellular networks. A connection switched from cellular networks to WLAN needs to up its bandwidth, otherwise it will lose the benefit of the integration. So we need to degrade some connections adaptively to accommodate more new arrivals and handoff calls. Some methods have been proposed [17–25] for this purpose. Our method tackles the problem from a new angle based on the concept of the proposed degradation profile. We effectively degrade the longest calls in the system based on their state information because they have a greater probability of quitting the system and leaving fewer degraded connections in the system. Use of the degradation profile can guarantee the satisfied QoS level to the end user and degrading the longest calls can reduce the degradation degree of the whole system. The pseudo-code of the adaptation algorithm is described in Table 4, where Bi represents the required bandwidth and Di denotes the minimum bandwidth request defined in the connection degradation profile. The level of the performance degradation of the overall system is critical information to the network operators. If this happens frequently in

178

X.G. Wang et al. / Computer Networks 47 (2005) 167–183

Table 4 Pseudo-code for the adaptation algorithm New call arrivals IF (New requested bandwidth Bi < system available bandwidth) assign Bi; ELSEIF (New requested band Di < system available bandwidth) assign Di; ELSE WHILE (undegraded call exists AND Di > system bandwidth) degrade longest call; IF (Di < system available bandwidth) assign Di; ELSE reject the call; Handoff call arrivals IF (Handoff requested bandwidth Bi < system available bandwidth + guard bandwidth) assign Bi; ELSEIF (Handoff requested band Di < system available bandwidth + guard bandwidth) assign Di; ELSE WHILE (undegraded call exists AND Di > system available bandwidth + guard bandwidth) degrade longest call; IF (Di < system available bandwidth+guard band) assign Di; ELSE reject the call; Departures WHILE (system available bandwidth > 0) find the shortest degraded call; assign Bi for this call;

a certain area, a new base station or a new access point may be installed to solve the problem permanently. For this purpose, we define a new performance merit called system degradation degree. Some system parameters are described before we introduce the definition of this concept. The traffic class of a connection is defined as Ci, where Ci 2 {C1, C2, . . ., Ci, . . ., CK}, where K is the number of service classes. The corresponding bandwidth requirement for each class is defined as Bi 2 {B1, B2, . . ., Bi, . . ., BK}, for the sake of simplicity we assume that all the connections in the same class have the same requested bandwidth. Di 2 {D1, D2, . . ., Di, . . ., DK} denotes the minimum bandwidth request defined in the connection degradation profile.

Let pi(t) denote the degradation probability of class i and ni(t) the number of connections from class i at time t. Thus the degradable bandwidth at time t can be written as K X

ðBi  Di Þpi ðtÞni ðtÞ:

ð3Þ

i¼1

We define bandwidth degradation degree BR as the ratio of the amount of reduced bandwidth and the requested bandwidth: PK ðBi  Di Þpi ðtÞni ðtÞ BR ¼ i¼1 PK : ð4Þ i i¼1 Bi n ðtÞ The overall system degradation degree SD is the integration of BR over the period t:

X.G. Wang et al. / Computer Networks 47 (2005) 167–183

SD ¼

Z PK

 Di Þpi ðtÞni ðtÞ : PK i i¼1 Bi n ðtÞ

i¼1 ðBi

t

ð5Þ

6. Performance analysis 6.1. The simulation model This section uses simulation experiments to investigate how the proposed approach can improve the overall QoS for the integrated cellular and WLAN networks. Following the assumptions widely used in previous studies [21,23,24], the call arrivals in our simulation follow an independent Poisson process and the session time of each connection is exponentially distributed. It is well known that dropping an established communication is worse than rejecting a new call. Therefore cellular systems reserve a guard bandwidth for the handoff calls in order to reduce the handoff dropping probability. The reserved guard bandwidth can be either static or dynamic [22,24,25]. The dynamic approach often outperforms the static one at the expense of generating more control overheads [3]. However, the static approach is often attractive in practice owing to its design simplicity. In our simulation, a static guard bandwidth (i.e., 5% of the system capacity) is employed to deal with handoff calls. The integrated network in the simulation consists of one cellular network and one WLAN hotspot. Since WLAN has a higher capacity and is cheaper than UMTS, we assume the handoff probability from UMTS to WLAN is 5 times as much as that from WLAN to UMTS. The system capacity for UMTS and WLAN is 2 and 11 mb/s respectively. The bandwidth requirement for each of four QoS classes {B1, B2, B3, B4} defined in Section 4.3 and their acceptable degradation level defined in degradation profile are assumed to be a portion of the system capacity listed in Table 5. The reservation signaling cost before the establishment of each new or handoff connection is set to a fixed value. For the sake of clarity, all the relevant

179

simulation parameters are summarized in Table 5. The simulation is carried out under various traffic loads. We compare the proposed approach with non-adaptive multimedia services. 6.2. The experiment results This section presents the simulation results to demonstrate the effectiveness of the proposed scheme. To do so, we assign the same traffic loads to two systems working under normal operation conditions and under the proposed framework. Each simulation experiment was run until the system reached its stable state. To measure the system performance merits, we first examine the normalized system utilization defined as the amount of data transmitted in the unit time normalized with the system capacity. We then consider QoS parameters: the call blocking probabilities and handoff dropping probabilities. Finally, the overall system degradation is calculated. Fig. 12 compares the bandwidth utilization supported by the proposed adaptive scheme in the integrated network to that without the adaptive scheme under various traffic loads. From this diagram, we can observe that the utilization increases as traffic loads increase. Under all system traffic loads, the adaptive strategy uses the system recourses more efficiently than non-adaptive connections. When the traffic load becomes higher, the advantage is more evident. The reason that adaptive connections can better utilize the system bandwidth is that the Table 5 Simulation parameters Parameter

Value

UMTS capacity (U) WLAN capacity (W) UMTS to WLAN handoff WLAN to UMTS handoff Reservation signaling cost Session time Guard band {B1, B2, B3, B4} {D1, D2, D3, D4} Simulation time

2 mb/s 11 mb/s 0.05 0.01 1% * W Exp(50) 5% {5% * W, 3% * W, 5% * U, 3% * U} {4% * W, 2% * W, 4% * U, 2% * U} 1000 s

180

X.G. Wang et al. / Computer Networks 47 (2005) 167–183

Utilization

0.9 0.8

Non-Adaptive

0.7

Adaptive

0.6 0.5 0.4 0.3 0.2 0.1 0 0

0.1

0.2

0.3

0.4 0.5 0.6 Traffic Load

0.7

0.8

0.9

Fig. 12. Utilization vs. traffic load.

proposed scheme allows the network to intelligently adjust each admitted QoS connection by its degradation profile and give sufficient resources for the new or handoff calls. Fig. 13 depicts the call blocking probability vs. the traffic load for adaptive connections and nonadaptive connections. From the diagram, we can observe that there is no call blocking probability for both methods with light traffic load. Particularly, we start to see the call blocking probability when the traffic loads reached 0.4 in the non-adaptive situation and for the adaptive conditions we start to observe the call blocking probability at 0.5 traffic load. This clearly demonstrates the effec-

tiveness of the proposed mechanism. With further increments of the traffic load, the call blocking probability increases, since channels became more and more crowded. The figure also reveals that the adaptive approach reduces the call blocking probability compared to the non-adaptive approach. Fig. 14 further evaluates the handoff dropping probability in the integrated network. From the figure, we can observe that the handoff dropping probabilities increase as the system traffic loads increase. The handoff dropping probability for adaptive connections is much less than that for non-adaptive connections at the same traffic load condition. When the traffic load becomes higher, the trend is more evident. For instance, when the rate of traffic loads reaches 0.9, the handoff dropping probability is 0.1079 for adaptive connections and 0.0104 for non-adaptive connections. Under the adaptation system, we barely see the handoff dropping calls and this only emerges at traffic load 0.7. This reveals that the proposed approach reduces a great number of handoff dropping calls for the integrated WLAN and cellular system, which is often a disturbing event in cellular networks. Fig. 15 shows the degree of overall system degradation defined in Section 5. This designed parameter can act as indicator to network operators. When the overall system degradation parameter stays high for a certain period time, the network operator should think of installing more base stations or access points. From the figure,

0.2

0.12 Non-Adaptive

Handoff Dropping Probability

Call Blocking Probability

0.18 Non-Adaptive

0.16

Adaptive

0.14 0.12 0.1 0.08 0.06 0.04 0.02 0 0.4

0.5

0.6

0.7

0.8

0.9

Traffic Load

Fig. 13. Call blocking probability vs. traffic load.

0.1

Adaptive

0.08 0.06 0.04 0.02 0 0.4

0.5

0.6 0.7 Traffic Load

0.8

0.9

Fig. 14. Handoff dropping probability vs. traffic load.

X.G. Wang et al. / Computer Networks 47 (2005) 167–183 0.045

System degrade degree

0.04 0.035 0.03 0.025 0.02 0.015 0.01 0.005 0 0.5

0.6

0.7

0.8

0.9

181

mechanism is also developed under the proposed QoS model to address the various challenges generated by designing integrated WLAN and 3G networks. The superior performance of the system is revealed via simulation experiments. The results show that the proposed scheme uses system resources efficiently. Simulation experiments also indicate that the adaptive multimedia framework outperforms the non-adaptive approach in terms of lower handoff dropping probability and call blocking probability while still maintain acceptable QoS for the end users.

Traffic Load

Fig. 15. System degrade degree over traffic load.

we observe that the degradation increases nearly linearly with the increment of traffic loads, since with more users wanting to use the channel, the system is more adaptive. Also in practical systems, this performance merit can be designed as a threshold for control of the system performance by the network operator.

7. Conclusions The rapid deployment of WLAN and the 3G cellular systems provide the two major technologies for future networks. The design of a network architecture that can efficiently integrate WLAN and cellular networks is a challenging task. Many difficulties emerge when providing QoS, such as the unbalanced capacity of the two systems, handoff due to user mobility and unreliable wireless media. To enable efficient use of the scarce resources provided by the cellular networks while also maintaining strong service guarantees, this study has proposed a generic reservation-based QoS model for the integrated cellular and WLAN networks. Our model supports the delivery of adaptive real-time flows for end users taking the advantage of high data-rate WLAN systems as well as the wide coverage area of cellular networks. In particular, we analyze the different components of the model and their interaction. An adaptation

References [1] L. Kleinrock, Nomadicity: anytime, anywhere in a disconnected world, Mobile Networks and Applications 1 (4) (1997) 351–357. [2] L. Kleinrock, On some principles of nomadic computing and multi-access communications, IEEE Communications Magazine 38 (7) (2000) 46–50. [3] D. Wisely, P. Eardley, L. Burness, IP for 3G, Wiley, New York, 2002. [4] K. Ahmavaara, H. Haverinen, R. Pichna, Integration of wireless LAN and 3G wireless––Interworking architecture between 3GPP and WLAN systems, IEEE Communications Magazine 41 (11) (2003) 74–81. [5] M. Welzl, M. Muhlhauser, Scalability and quality of service: a trade-off ? IEEE Communications Magazine 41 (6) (2003) 32–36. [6] IEEE. std., Wireless LAN Medium Access Control (MAC) and Physical Layer (PHY) Specifications, ISO/IEC 880211:1999(E), 1999. [7] IEEE. 802.11e/D4.3, Draft Supplement to STANDARD FOR Telecommunications and Information Exchange Between Systems––LAN/MAN Specific Requirements–– Part 11: Wireless Medium Access Control (MAC) and Physical Layer (PHY) specifications: Medium Access Control (MAC) Enhancements for Quality of Service (QoS), 2003. [8] S. Mangold, S. Choi, P. May, O. Klein, G. Hiertz, L. Stibor, IEEE 802.11e wireless LAN for quality of service, in: European Wireless Conference, 2002, pp. 32–39. [9] S. Choi, J. Del Prado, S. Shankar, N.S. Mangold, IEEE 802.11e contention-based channel access (EDCF) performance evaluation, in: 2003 International Conference on Communications (ICC 2003), May 11–15, 2003, 2003, pp. 1151–1156. [10] 3GPP, Group Service and System Aspects; 3GPP Systems to Wireless Local Area Network (WLAN) Interworking; System Description, R. TS 23.234, Ed., May 2003. [11] GPP, End-to-End QoS Concept and Architecture, R. TS 23.207, Ed., January 2002.

182

X.G. Wang et al. / Computer Networks 47 (2005) 167–183

[12] S. Dixit, Y. Guo, Z. Antoniou, Resource management and quality of service in third generation wireless networks, IEEE Communications Magazine 39 (2) (2001) 125–133. [13] S.I. Maniatis, E.G. Nikolouzou, I.S. Venieris, QoS issues in the converged 3G wireless and wired networks, IEEE Communications Magazine 40 (8) (2002) 44–53. [14] ETSI, Broadband Radio Access networks (BRAN); HIPERLAN Type 2; Requirements and Architectures for Interworking between HIPERLAN/2 and 3rd Generation CellularSystems, TR 101 957. [15] X.G. Wang, J.E. Mellor, K. Al-Begain, G. Min, Supporting QoS in integrated IP-based cellular and WLAN networks using an adaptive framework, in: First International Working Conference on Performance Modelling and Evaluation of Heterogeneous networks (HETNETs Õ03), 2003. [16] T.P. Barzilai, D.D. Kandlur, A. Mehra, D. Saha, Design and implementation of an RSVP-based quality of service architecture for an integrated services Internet, IEEE Journal on Selected Areas in Communications 16 (3) (1998) 397–413. [17] X. Wang, H. Schulzrinne, Integrated resource negotiation, pricing and QoS adaptation framework for multimedia applications, IEEE Journal on Selected Areas in Communications 18 (12) (2000) 2514–2529. [18] M. Naghshineh, M. Willebeek-LeMair, End-to-end QoS provisioning in multimedia wireless/mobile networks using an adaptive framework, IEEE Communications Magazine 35 (11) (1997) 72–81. [19] M. Mirhakkak, N. Schult, D. Thomson, Dynamic bandwidth management and adaptive applications for a variable bandwidth wireless environment, IEEE Journal on Selected Areas in Communications 19 (10) (2001) 1984–1997. [20] F. Prihandoko, M.H. Habaebi, B.M. Ali, Adaptive call admission control for QoS provisioning in multimedia wireless networks, Computer Communications 26 (14) (2003) 1560–1569. [21] M. El-Kadi, S. Olariu, H. Abdel-Wahab, A rate-based borrowing scheme for QoS provisioning in multimedia wireless networks, IEEE Transactions on Parallel and Distributed Systems 13 (2) (2002) 156–166. [22] P. Ramanathan, K.M. Sivalingam, P. Agrawal, S. Kishore, Dynamic resource allocation schemes during handoff for mobile multimedia wireless networks, IEEE Journal on Selected Areas in Communications 17 (7) (1999) 1270–1283. [23] C.L.P. Chen, Y. Xiao, B. Wang, Bandwidth degradation QoS provisioning for adaptive multimedia in wireless/ mobile networks, Computer Communications 25 (13) (2002) 1153–1161. [24] C. Oliveira, J.B. Kim, T. Suda, Adaptive bandwidth reservation scheme for high-speed multimedia wireless networks, IEEE Journal on Selected Areas in Communications 16 (6) (1998) 858–874. [25] S. Choi, K.G. Shin, Adaptive bandwidth reservation and admission control in QoS-sensitive cellular networks, IEEE Transactions on Parallel and Distributed Systems 13 (9) (2002) 882–897.

Xin Gang Wang received his first B.Sc. degree in Computer Science from the Heilongjiang University, P.R. China, in 2001. He is currently a Ph.D. student in the computing department, University of Bradford. His research interests include performance modeling of mobile networks.

Geyong Min received the Ph.D. degree in computing science from the University of Glasgow, United Kingdom, in 2003, and the B.Sc. degree in computer science from Huazhong University of Science and Technology, China, in 1995. He is currently a lecturer in the Department of Computing at the University of Bradford, United Kingdom. His research interests include performance modelling/evaluation, parallel and distributed systems, mobile computing, computer networks, multimedia systems. He is the founding co-chair of the International Workshop on Performance Modelling, Evaluation, and Optimisation of Parallel and Distributed Systems (PMEO-PDS) held in conjunction with IEEE/ACM-IPDPS. He is the guest editor of the journals Computation and Concurrency: Practice and Experience, Future Generation Computer Systems, and Supercomputing. He has served on the program committees of a number of international conferences. He is a member of the IEEE Computer Society.

John Mellor has worked in the modelling and simulation of communication networks for 25 years. Early work included dynamic alternate routing and the application of learning automata to routing strategies in circuit and packet switched networks. Collaboration with a Cambridge UK company led to the development of a LAN protocol which consistently outperformed Ethernet. He was sent as a government expert to study the manufacturing messaging protocol in the USA and Japan. He later became a technical expert consultant on the application of European Directives within the manufacturing industry. A forray into radio frequency identification tags resulted in the development of a novel protocol that was exploited by a major vehicle component manufacturer. He now finds himself involved in wireless protocols with researchers working on WiFi (802.11) and on security aspects of mobile commerce. He is leader of the Mobile Computing and Networks Research Group at the University of Bradford and course tutor to three innovative advanced M.Sc. courses in mobile computing, applications and security.

X.G. Wang et al. / Computer Networks 47 (2005) 167–183 Khalid Al-Begain is Professor of Mobile Networking and Head of the Mobile Computing and Networking Research Centre at the School of Computing of the University of Glamorgan in Cardiff/Wales/UK. He received his High Diploma (1986), the Specialisation Diploma of Communication Engineering (1988) and his Ph.D. degree in Communication Engineering (1989) from the Technical University of Budapest in Hungary. From 1990 to 1996, he held the position of a Assistant Professor at the Department of Computer Science of the MuÕtah University in Jordan. Then he became an Associate Professor at the same university. In 1997 he moved to the Department of Computer Science at the University of Erlangen-Nuremberg in Germany as Alexander von Humboldt research fellow. Later, he spent one year as Guest Professor at the Chair of Telecommunications, Dresden University of Technology, Germany. From 2000 to 2003, he has been Senior Lecturer and Director of Postgraduate Research in the Department of Computing of the University of Bradford, UK before moving to Glamorgan. He coauthored the book ‘‘Practical Performance Modelling’’ published by Kluwer Academic Publishers in Boston and more than 100 refereed journal and conference papers. He also served/serves as Guest Editor for several special issues of the International Journal of Simulation on Analytical and

183

Stochastic Modelling Techniques. He is UNESCO Expert in networking, UK Representative to EU COST Action 290 Management Committee, senior member of the IEEE and many other scientific organisations. Since 2003, he is the Conference Chair for the annual ASMTA (Analytical and Stochastic Modelling Techniques and Applications) Conference (ASMTAÕ03 in Nottingham, UK and ASMTAÕ04 in Magdeburg, Germany). He also manages several research projects funded by the EPSRC and EU. His research interests include performance modelling and analysis of computer and communication systems, modelling and design of wireless mobile networks and multicast routing in mobile IP networks. He is also interested in mobile computing research. Lin Guan received the B.Sc. degree in computer science from Heilongjiang University, Heilongjiang, China, in 2001. She is currently a Ph.D. student in University of Bradford. Her research interests focus on developing cost effective analytical models for the performance evaluation of congestion control algorithms for Internet traffic.