emerging wireless multimedia services and technologies phần 6 docx

To this purpose, we formulate a number of practical interworkingscenarios, where UMTS subscribers with ongoing real-time video sessions handover to WLAN, and weconsider the capability of

(5) Assuming a QSTAihaving niactive TSs is selected for polling, the scheduler calculates the TXOPduration TDi, in three steps:

(a) First, for every TSij of QSTAið j ¼ 1 niÞ, the scheduler calculates TDij, as the maximum

of (i) the time required to accommodate the pending traffic, as indicated by the queue size ofthat TSðQSijÞ, plus any overheads and, (ii) mTDij, to ensure that the assigned TXOP will have

at least the minimum duration:

TDij¼ max QSij

Rij þ O; mTDij

In the special case where QSijis equal to zero, TDijis set equal to the time for the transmission

of a Null-Data MSDU, to allow QSTAito update the queue size information for TSij.(b) TDifor QSTAiis calculated as the sum of all TDij:

TDi¼Xn i j¼1

of the algorithm, we refer to ‘basic ARROW’ and ‘enhanced ARROW’ The idea behind the enhancement

is that, instead of waiting for the QS information or use Null-Data TXOPs to get the current queue size,the scheduler can estimate the current queue size of a CBR TS Every time a TXOP is assigned to aQSTA with CBR TSs, the scheduler calculates the TXOP duration for each of these TSs by adding thequeue size value, indicated by the previous MSDU transmission of the same TS, and the estimated(using Mean Data Rate) generated traffic in the time interval between the previous and the currenttransmission Accordingly, for CBR TSs Equation (7.11) can be replaced by:

TDij¼ max QSijþ

ij

t t0

no expense to cover cases when, for example, a MSDU generation is delayed for some reason (e.g., highcomputing load), and misses the scheduled TXOP Assuming a CBR TS in Figure 7.6, the duration ofTXOPiðx þ 1Þ is calculated based on the QSiðxÞ and the estimation for the generated MSDUs within theinterval½tðxÞ; tðx þ 1Þ This estimation is very accurate for CBR TSs, leading to considerably lowertransmission delays, since the MSDUs generated in the interval ½tðxÞ; tðx þ 1Þ are transmitted attðx þ 1Þ, instead of tðx þ 2Þ with basic ARROW This strategy also reduces transmission overheads andleads to lower average channel occupancy (i.e., better channel utilization) since, by picking a suitablevalue for mSIi, adequate for the generation of at least one MSDU, no Null-Data TXOPs for CBR TSs arerequired.

To show how traffic scheduling can affect the performance of 802.11e, simulation results from thecomparison of Simple, SETT-EDD, basic ARROW and enhanced ARROW are presented below Thesimulation scenarios considered an increasing number of QSTAs attached to an AP All QSTAs and the

AP were supporting the extended MAC layer specified in IEEE 802.11e [17] and the PHY layerspecified in IEEE 802.11g [18], with a transmission rate of 12 Mbps An ideal, error-free wirelesschannel was assumed, as the focus was on the scheduling procedure In order to investigate the limitsand the maximum scheduling capability of each algorithm under heavy traffic conditions, no admissioncontrol was applied Additionally, no minimum fraction of time for the operation of EDCA was reserved(i.e., the whole Beacon Interval could be utilized by HCCA) to focus solely on the performance of theHCCA mechanism Each QSTA had two active sessions:

(1) a bi-directional G.711 voice session (CBR traffic), mapped into two TSs (one per direction), and(2) an uplink (from QSTA to AP) H.261 video session at 256 Kbps (VBR traffic), mapped into oneuplink TS

Figure 7.7 depicts throughput of non-delayed MSDUs for voice and video traffic For voice traffic(Figure 7.7(a)), basic ARROW accommodates up to 18 QSTAs, while SETT-EDD can manage up to 14QSTAs and Simple up to only 7 QSTAs Using the enhancement, the number of QSTAs can be increased

to 19 with enhanced ARROW For video traffic (Figure 7.7(b)), basic and enhanced ARROW outperformboth SETT-EDD and Simple, accommodating up to 19 QSTAs, as opposed to 13 with SETT-EDD and 6with Simple The main reason for the considerably improved performance of basic ARROW is theaccurate TXOP assignment it performs, as a result of the accurate queue size information This is alsoshown in more detail later in this section As for the enhanced ARROW, it appears that the admissioncapacity limit of the Scheduler compared with the Standard ARROW for Voice traffic is somewhatincreased since, with Enhanced ARROW up to 19 G.711 TSs can be accommodated comfortably

It is interesting to observe that throughput of SETT-EDD and ARROW (both basic and enhanced)reduces rapidly immediately after reaching its maximum value The reason is that, due to the dynamicTXOP assignment performed by these algorithms, new TSs entering the system can participate equally

in the channel assignment Thus, when the scheduler exceeds its maximum scheduling capability,service for all TSs is degraded abruptly The Simple Scheduler on the other hand, manages to provide astable throughput regardless of the offered load, because static allocations for existing TSs are notaffected as the traffic load increases This effect highlights the need for an effective admission controlscheme for SETT-EDD and ARROW, which would prevent the offered load from exceeding themaximum scheduling capability

In Figure 7.8, the mean delay for non-delayed voice and video MSDUs is shown For voice traffic(Figure 7.8(a)), both Simple and SETT-EDD perform better than basic ARROW, since they are able toperform a fairly accurate estimation of expected CBR traffic But enhanced ARROW achievesconsiderable improved delays, comparable to both Simple and SETT-EDD, as a direct effect of theincorporation of the traffic load prediction for CBR traffic For video traffic (Figure 7.8(b)), Simpleachieves a low and almost stable mean delay, but suffers from the considerable low number ofaccommodated QSTAs, as shown earlier in Figure 7.7(b) The improved delay of basic ARROW

compared with SETT-EDD shows that the extra scheduling delay introduced by ARROW is balanced bythe accurate TXOP assignment With the enhanced version of ARROW, the mean delay for video traffic

is the same as with basic ARROW for low traffic loads (up to seven QSTAs), but for higher traffic loadsthe delay is somewhat increased This occurs because of the use of the estimation mechanism thateliminates the need for Null-Data TXOPs for CBR traffic and results in the allocation of larger TXOPsfor voice traffic, compared with basic ARROW This delays the TXOP allocation for video MSDUs andincreases their mean delay in heavy traffic conditions But, in any case, the transmissions are within thedelay bounds of video traffic

G.711 Non-Delaye d M SDU Throughput

(b) H.261 videoFigure 7.7 Throughput of non-delayed MSDUs.

7.3 RSVP over Wireless Networks

QoS provision is considered today as one of the basic requirements for next generation networks Thesenetworks are expected to integrate a large number of access systems, both wired and wireless, in onecommon core, and rely on the Internet Protocol (IP) for data exchange The IP protocol, up to its currentversion 4, was designed for applications with low network requirements, such as e-mail and ftp, andaccordingly offers an unreliable service that is subject to packet loss, reordering, packet duplication andunbounded delays The expected new version 6 of IP provides some means for QoS support, but it still

G.711 M ean Delay of ND M SDUs

(b) H.261 videoFigure 7.8 Mean delay of non-delayed MSDUs.

needs supporting mechanisms to provide guaranteed service To treat the problem of QoS in IPnetworks, the Internet Engineering Task Force (IETF) has introduced two main frameworks, namely theIntegrated Services (IntServ) [1] and the Differentiated Services (DiffServ) [18] DiffServ classifies andpossibly conditions the traffic, in order to ensure similar behaviour throughout the network It performswell in core networks, due to its scalability to support large numbers of flows IntServ on the other hand,

is targeted mainly for the access systems, by providing means to request and obtain end-to-end QoS perflow As already mentioned, the Resource Reservation Protocol (RSVP) [2] is considered the mostpopular signaling protocol in IntServ for requesting QoS per flow, and setting up reservations end-to-endupon admission

A static resource reservation based approach (such as RSVP) performs well in fixed networks wherelinks are stable, but exhibits low performance in a variable bandwidth environment RSVP assumes astable given overall bandwidth per link, that can be distributed to different active flows, based on theirrequirements When a new flow requests admission, RSVP is used to check if the overall bandwidth perlink is sufficient for all (active and new) flows If the result of this check is positive, the new flow can beaccepted But if the available bandwidth is reduced after admission control (as can be the case inwireless links), the network may not be able to meet commitments for flows that have been successfullyadmitted In this case, none of the flows operates according to its requirements, while the QoS observed

by the user can be poor

A number of solutions can be found in the literature, addressing this problem For example, dynamicRSVP (dRSVP) [19] uses ranges of traffic flows specifications, instead of absolute values The benefit isthat intermediate nodes can adjust allocations, as long as these are within the specified ranges,depending on changes of the available bandwidth in the output links As expected, dRSVP introduces

a number of extensions/modifications of dRSVP, compared with the standard RSVP as listedbelow

An additional flow specification object (FLOWSPEC) in RESV messages and an additional trafficspecification object (SENDER TSPEC) in PATH messages have been introduced, so as to describeranges of traffic flows

A measurement specification object (MSPEC) has been added in the RESV messages, used to allownodes to learn about downstream resource bottlenecks

A measurement specification object (SENDER MSPEC) has been introduced in the PATH messages,used to allow nodes to learn about upstream resource bottlenecks

An admission control process has been added, able to handle ranges of required bandwidth

A bandwidth allocation algorithm has been introduced that divides the available bandwidth amongadmitted flows, taking into account the desired range for each flow, as well as any upstream ordownstream bottlenecks

Measurements presented in [19] have shown that dRSVP can significantly improve the performance ofRSVP over wireless Nevertheless, as listed above, it introduces a number of enhancements to thestandard protocol Another solution that does not introduce changes to the protocol operation is to detectthe instances where the overall bandwidth is reduced below the required limit and reject a number offlows in order to allow the rest to operate as required The question is how many and which in particularshould be rejected, to cause as little inconvenience to the users and the network as possible Below, wepresent such a mechanism for fair and efficient flow rejection More information on this mechanism can

be found in [20]

The presented flow rejection mechanism can operate at a central point of the wireless link (e.g., theAccess Point of a cell), where it can observe the overall provided bandwidth When the providedbandwidth falls to a level that is insufficient for maintaining the reserved bandwidth values for all flows,one or more flows must be rejected, in order to maintain committed values for the rest This can occur intwo cases

(1) When a new flow requests admission If bandwidth is insufficient, the admission control will notallow the new flow to be accepted.

(2) When the overall available bandwidth is decreased to a point that the requested bandwidth per flowcannot be maintained for all flows In this case, one or more of the flows have to be rejected, inorder to maintain the reserved bandwidth for the rest of the flows within limits

Flow rejection can be performed through standard RSVP signaling (RESV_Tear, PATH_Tear) A simplemechanism can be used that rejects flows randomly until the total requested bandwidth is reduced belowthe available capacity This can lead to low efficiency, in terms of flow rejection probability andbandwidth utilization, as it might tear down:

more flows than necessary (this can happen when a large number of flows with low bandwidthrequirements is randomly selected for rejection, instead of a smaller number of high bandwidthflows),

high priority flows (if the mechanism does not differentiate the flows, based on their importance, itcan reject high priority instead of low priority flows), or

flows that utilize a large portion of the available bandwidth (this could lead to low bandwidthutilization)

Below, a more efficient flow rejection mechanism is presented, which attains a considerably improvedflow rejection probability In this, three criteria are considered for providing fair and efficientrejection

(1) Reject the lower priority flows first In order to achieve this, a scheme is needed that classifies flowsinto a number of priority classes The priority classes may be defined according to a number ofcriteria, such as the flow type, traffic characteristics, QoS requirements, etc In its simplest form, theclassification scheme can use the transport-layer protocol type included in every session identifica-tion triplet to classify flows For example, UDP flows can be considered as high priority, as theyusually carry real time data, while TCP flows can be considered as low priority

(2) Minimize the number of rejected flows If there are more that one set of flows that can be rejected,then the set with the fewer members (i.e., flows) should be selected for rejection This criterionprevents one from rejecting a large number of low bandwidth flows

(3) Prevent underutilization This could happen if the mechanism chose to reject flows with largebandwidth requirements Accordingly, the mechanism should reject a set of flows that leaves a totalrequired bandwidth lower than but as close to the available capacity as possible

More specifically, the mechanism takes effect when the available bandwidth of the wireless link fallsbelow the total required bandwidth of all flows Assuming an ascending list of flowsffi ;1; fi ;2; ; fi ;ng ineach priority class Ci, according to their bandwidth requirements, the mechanism starts with the firstflow f1 ;1 of the lowest priority list C1, and checks if, by rejecting it, the total required bandwidth fallsbelow the available It continues traversing the list until:

(1) either such a flow is found, or,

(2) the end of the list is reached

In case (1), it rejects the flow and stops, since the total required bandwidth is below the available In case(2), it rejects the last flow of the list f1 ;n(the one with the maximum bandwidth reservations in the list)and starts over from the beginning of the list If all of the flows in the list are rejected (i.e., all the flows

of the particular priority class), it continues with the flows of the next higher priority class C2, and so on,until the total required bandwidth falls below the available, or until all flows have been rejected

Below, the results of a comparison between the previously presented mechanism and a randomrejection mechanism are presented, based on a simulation model with the following characteristics.(1) Flows belong to two different priority classes, C1¼ Low and C2¼ High, although the mechanismcan work with any number The mechanism starts dropping flows belonging to priority class Low,

as previously described in detail

(2) Each priority class includes a mix of flow types with different bandwidth requirements, assumingthat the type of a flow is an independent discrete random variable

(3) The number of active flows is not constant In each class, flows arrive according to an independentPoisson process Flow durations are assumed to be exponential independent random variables.(4) The channel can be found in two states, ‘good’ or ‘bad’, offering two levels of finite bandwidth Thetime the channel stays in each state is an exponential random variable, thus the channel can bemodeled as a two state Markov chain

(5) New flows are not allowed to enter the system when bandwidth is insufficient to satisfy therequirements of all (existing and new) flows

Two simulation scenarios were considered, where flows in each class were generated as in Table 7.1 Ascan be observed, the flow mix included a large number of ‘light’ flows and a smaller number of ‘heavy’flows To experiment with different conditions, we used different channel capacities and flow interarrivaltimes per class In both scenarios, the rejection probability per class was measured, defined as theaverage number of flows dropped divided by the total number of flows in the class

Scenario 1

Here, a low capacity channel was considered, with bandwidth equal to 600 and 400 while in the ‘good’and ‘bad’ state, respectively The mean dwell times in each state were 8 and 1, respectively The meanflow duration was equal to 4 for all flows, while six experiments were performed with differentinterarrival times per class in the range [0.012, 0.040] As can be seen in Figure 7.9, the proposedmechanism attains lower overall rejection probability in all experiments, while for the high priority classthe performance is significantly improved The increased rejection probability for the low priority class

is considered acceptable, since this class is destined mostly for best-effort flows

Scenario 2

In this scenario, the channel capacity was increased by a factor of 4 (Good¼ 2400, Bad ¼ 1600), whilethe range of interarrival times was decreased by the same factor ([0,003, 0,009]), resulting in aconsiderably larger number of active flows The mean time per channel state and the mean flow durationwere as in Scenario 1 Six experiments were performed with different interarrival times per class, andthe results are presented in Figure 7.10 Again, the overall rejection probability is improvedsignificantly, especially for large values of the interarrival time The rejection probability for the highpriority class is much lower than the corresponding mean value of the randomly rejection mechanism

Table 7.1 Different bandwidth requirements per class

Flow types Bandwidth requirements Probability

In both scenarios, the channel utilization of the proposed mechanism was equal or slightly improvedthan that for the random rejection mechanism, indicating that the proposed mechanism manages toreduce the flow rejection probability without decreasing the bandwidth utilization.

The interworking between 3G cellular and Wireless Local Area Networks (WLANs) has beenconsidered as a suitable and viable evolution path towards the next generation of wireless networks

Figure 7.10 Rejection probability vs mean interarrival time (high offered traffic scenario).

b Parts of the following sections have been published in [34].

Figure 7.9 Rejection probability vs mean interarrival time (low offered traffic scenario).

Yet this interworking raises considerable challenges, especially when we demand seamless continuity ofmultimedia sessions across the two networks To deal with these challenges several 3G/WLANinterworking requirements need to be identified and fulfilled.

Typically, the 3G/WLAN interworking requirements are specified and categorized in terms of severalusage scenarios [22, 23] For example, a common usage scenario is when a 3G subscriber is admitted to

a WLAN environment by re-using his regular 3G credentials, and then obtains an IP connectivityservice (e.g., access to the Internet) In this case, the interworking requirements include support of 3G-based access control, signaling between the WLAN and the 3G network for Authentication, Authoriza-tion and Accounting (AAA) purposes, etc Other scenarios can call for more demanding interworkingrequirements We may envisage, for instance, a scenario in which a 3G subscriber initiates a videosession in his home 3G network and subsequently transits to a WLAN environment, wherein the videosession is continued seamlessly, i.e., without any noticeable change to the quality of service (QoS) Inthis case, not only 3G-based access control is required, but also access to 3G-based services is neededover the WLAN network, which in turn calls for appropriate routing enforcement mechanisms Moreimportantly, however, there is need for QoS consistency across 3G and WLAN, which does not appear

to be very straightforward given the different QoS features offered by these networks Indeed, WLANshave initially been specified without much attention being paid to QoS aspects and aimed primarily tosatisfy simple and cost-effective designs Even with the recent IEEE 802.11e [17] developments,WLAN QoS still exhibits several deficiencies with respect to the 3G QoS (this is further discussed later)

On the contrary, 3G cellular networks were built with the multimedia applications in mind and tradesimplicity and cost for inherently providing enhanced QoS in wide-area environments

Our main interest in the following sections of this chapter is to examine the challenges of seamlesssession continuity across UMTS and WLAN, and to evaluate the conditions and restrictions under whichseamless continuity is feasible To this purpose, we formulate a number of practical interworkingscenarios, where UMTS subscribers with ongoing real-time video sessions handover to WLAN, and weconsider the capability of WLAN to provide seamless session continuity under several policy rules andWLAN traffic loads One measure that we are particularly interested to quantify is the maximumnumber of UMTS subscriberscthat can be admitted to the WLAN, subject to maintaining the level ofUMTS QoS and respecting the WLAN policies Although our study focuses primarily on QoSconsistency, we do address several other issues that are equally important for enabling seamless sessioncontinuity, such as routing enforcement, access control and differentiation between the traffic of regularWLAN data users and UMTS roamers The framework for discussing these issues is created byconsidering a practical UMTS/WLAN interworking architecture that conforms to the 3GPP specifica-tions [23, 24] and other interworking proposals found in the technical literature (e.g., [22])

7.5 UMTS/WLAN Interworking Architecture

The end-to-end interworking architecture we are considering is illustrated in Figure 7.11 and iscompliant with the proposals in [22] and [23] Below we briefly discuss the main characteristics ofthis architecture Although we do not provide a comprehensive description, we set the ground for thenext sections and define the real-life environment to which the subsequent study applies This creates theright context for the following discussion and makes it easy to assess the importance of the providedresults For more detailed information on the considered architecture the interested reader is referred to[22–26]

As shown in Figure 7.11, the 3G network supports access to a variety of IP multimedia services viatwo radio access technologies: UMTS Terrestrial Radio Access (UTRA) and WLAN access Access

c The UMTS subscribers admitted to the WLAN are also referred to as UMTS roamers.

control and traffic routing for 3G subscribers in UTRA is handled entirely by the UMTS Switched (PS) network elements, which encompass the Serving GPRS Support Node (SGSN) and theGateway GPRS Support Node (GGSN) (see [25] for more details) On the other hand, access control andtraffic routing for 3G subscribers in WLAN (UMTS roamers) is shared among the WLAN and theUMTS network elements as discussed below The important assumption we make, as shown inFigure 7.11, is that 3G subscribers can change radio access technology and keep using their ongoingmultimedia sessions in a seamless fashion Thus, we assume that seamless service continuity isprovided This assumption raises considerable challenges and, as noted above, we are interested ininvestigating the capability of WLAN to support this seamless service continuity under specificscenarios Note that by the term seamless service continuity we to refer to the continuation of a session(that has been initiated in the UMTS environment) in the WLAN environment without any noticeableQoS degradation or any other discrepancies whatsoever We do not take into account however, thehandover latency In other words, we focus on ‘how well’ a session is continued in the WLAN and not

Packet-‘how fast’ the session is handed over from UMTS

The WLAN access network may be owned either by the UMTS operator or by any other party (e.g., apublic WLAN operator, or an airport authority), in which case the interworking is enabled and governed

by appropriate business and roaming agreements As shown in Figure 7.11, in a typical deploymentscenario the WLAN network supports various user classes, e.g., UMTS roamers and regular WLANdata users (i.e., no 3G subscribers) Differentiation between these user classes and enforcement ofcorresponding policies is typically enabled by employing several Service Set Identifiers (SSIDs) Forexample, the regular WLAN data users may associate with the SSID that is periodically broadcast by theAccess Point (AP), whereas the UMTS roamers may associate with another SSID that is also configured

in the AP, but not broadcast (see [26] about the usage of SSIDs) In this case, the WLAN can applyFigure 7.11 The considered end-to-end interworking architecture for seamless multimedia session continuity.

distinct access control and routing policies for the two user classes and can forward the traffic of WLANdata users, e.g to the Internet and the traffic of UMTS roamers to the UMTS PS core network (as shown

in Figure 7.11) Such routing enforcement is vital for supporting seamless service continuity and can beimplemented as discussed in [22] Moreover, different AAA mechanisms could be used for the differentuser classes

For enabling interworking with WLANs, the UMTS PS core network incorporates three newfunctional elements: the 3G AAA Server, the WLAN Access Gateway (WAG) and the Packet DataGateway (PDG) The WLAN needs to also support similar interworking functionality to meet the accesscontrol and routing enforcement requirements The 3G AAA Server in the UMTS domain terminates allAAA signaling originated in the WLAN that pertains to UMTS roamers This signaling is securelytransferred across the Wr/Wb interface, which is typically based on Radius [14] or Diameter [27]protocols The 3G AAA Server interfaces with other 3G components, such as the PDG, the HomeSubscriber Server (HSS), the Home Location Register (HLR), the Charging Gateway/ChargingCollection Function (CGw/CCF) and the Online Charging System (OCS) Note that, for simplicity,not all these elements are shown in Figure 7.11 Both the HLR and the HSS are basically subscriptiondatabases, used by the 3G AAA Server for acquiring subscription information for particular WLANMSs Typically, if an HSS is available, the HLR need not be used The CGw/CCF and the OCS are 3Gfunctional elements used to provide off-line and on-line charging services respectively The 3G AAAServer can also route AAA signaling to/from another 3G networks, in which case it serves as a proxyand it is referred to as 3G AAA Proxy (see [22])

As shown in Figure 7.11, traffic from UMTS roamers is routed to the WAG across the Wn interfaceand finally to the PDG across the Wp interface This routing is enforced by establishing appropriatetraffic tunnels after a successful access control procedure The PDG functions much like a GGSN in aUMTS PS core network It routes the user data traffic between the MS and an external Packet DataNetwork (PDN) (in our case, the IP Multimedia Network) and serves as an anchor point that hides themobility of the MS within the WLAN domain The WAG functions mainly as a route policy element,i.e., it ensures that user data traffic from authorized MSs is routed to the appropriate PDGs, locatedeither in the same or in a foreign UMTS network

7.5.1 Reference Points

Some of the reference points (also referred to as interfaces) illustrated in the architecture diagram ofFigure 7.11, are briefly discussed below A more thorough discussion of all reference points can befound in [23] A brief description of the 3G internal interfaces, such as Gr, Iu-ps, Gi, Gn, can be found in[25]

Wr/Wb This interface carries AAA signaling between the WLAN and the 3G visited or home PLMN

in a secure manner The proposed protocol across this interface is Diameter [27], which is used forproviding AAA functions and for carrying the Extensible Authentication Protocol (EAP) messagesexchanged between the MS and the 3G AAA Server However, since legacy WLANs need to besupported, the Radius [14] protocol must also be supported across Wr/Wb and this calls for Radius–Diameter interworking functions across the legacy WLAN and the 3G AAA Proxy/Server Diameter

is the preferred AAA protocol since it can provide enhanced functionality compared with Radius.Note that the term Wr is used to refer specifically to the interface that carries authentication andauthorization information, whilst the Wb is used to refer to the interface that carries accountinginformation

Wf This is located between the 3G AAA server and the 3G CGw/CCF (not shown in Figure 7.11).The prime purpose of the protocols on this interface is to transport charging information towards the3G operator’s CGw/CCF located in the visited or home PLMN The application protocol on thisinterface is Diameter based

Wo This interface is used by the 3G AAA server to communicate with the 3G OCS and exchangeonline charging information so as to perform credit control for the online charged subscribers Theapplication protocol across this interface is again Diameter based.

Wx This interface is located between the 3G AAA Server and HSS, and is used primarily foraccessing the WLAN subscription profiles of the users, retrieving authentication vectors, etc Asnoted before, if Wx is implemented, then the interface to the HLR is not required The protocolacross Wx is either Mobile Application Part (MAP) or Diameter-based

7.5.2 Signaling During UMTS-to-WLAN Handover

Although Figure 7.11 shows the architecture that can support seamless session continuity, it does notaddress the dynamics of handover procedure, which is especially important for the provision of seamlesscontinuity To elaborate further on this key procedure, we depict in Figure 7.12 a typical signalingdiagram that pertains to a situation where a mobile station (MS) hands over from UMTS to WLAN inthe middle of an ongoing packet-switched video session The establishment of the video session istriggered at instant A and, in response, the MS starts the Packet Data Protocol (PDP) contextestablishment procedure for requesting the appropriate QoS resources (described by the ‘Req QoS’Information Element (IE), see [33]) The UMTS network acknowledges the request and indicates thenegotiated QoS resources (specified by the ‘Neg QoS’ IE) that could be provided After that, videotraffic on the user plane commences and the video session gets in progress At some point the MS enters

a WLAN coverage area and it starts receiving Beaconsd from the nearby Access Points (APs) We

Activate PDP Context Request (Req QoS IE) Activate PDP Context Accept (Neg QoS IE)

UMTS QoS Resources Reserved

Beacon (SSID(b), QoS Capability)

Open System Authentication Association Request (SSID(g), QoS Capability)

WLAN QoS Resources Reserved

Association Response (Status)

A

Initiate a packet-switched

UMTS-based Access Control and Tunnel Establishment

Probe Response (SSID(g)) Probe Request (SSID(g)) Attach to UMTS PS Domain

C Decision to handover to WLAN

Figure 7.12 Typical signaling during handover of a video session from UMTS to WLAN (HCCA availability is assumed).

d From the Beacons the MS discovers what particular QoS features the WLAN supports, if any.

assume that this can happen concurrently with the ongoing video session because, although the MS hasone transceiver available, it can periodically decode signals on other frequency channels for inter-system handover purposes The MS may need to check if the detected WLAN supports one of itspreferred Service Set Identifiers (SSIDs) before considering it valid for inter-system change For thispurpose, the MS probes for a preferred SSID, denoted as SSID(g), according to the applicableprocedures in [26].

At instant C the MS takes the decision to handover to the detected WLAN and thus suspends theongoing video session This may demand further signaling with the UMTS but here we skip this forsimplicity After switching to the WLAN channel, the normal 802.11 authentication and associationprocedures [26] are carried out Subsequently, the UMTS-based access control procedure is executed inwhich the MS is authenticated and authorized by means of its regular 3G credentials (see [22], [23] formore details) At this stage, a tunnel will also be established for routing further MS traffic to a UMTSentry point (the WAG according to Figure 7.11) Next, the MS uses 802.11e QoS signaling (assuming it

is supported by the WLAN) to reserve the appropriate resources for its suspended video session TheTraffic Specification (TSPEC) element carries a specification of the requested QoS resources For theobjectives of seamless continuity, it is apparently that TSPEC needs to be set consistently with the QoSnegotiated in the UMTS system After this point, the video session is finally resumed in the WLAN,possibly after some high-layer mobility management procedures (e.g., Mobile IP or SIP)

From the above discussion it becomes evident that vertical handovers from UMTS to WLAN (andvice versa) present several challenges, especially for minimizing the associated latencies and theinterruption of ongoing multimedia sessions Apart from that however, the maintenance of consistentQoS across the UMTS and WLAN networks is equally challenging and is the focus of our discussionbelow

7.6 Interworking QoS Considerations

One vital component for the provision of seamless multimedia session continuity is the QoS consistencyacross the WLAN and UMTS networks This is indeed vital because without QoS consistency themultimedia sessions would experience different QoS levels in the two network domains and henceseamless continuity would not be doable It is unfortunate, however, that the UMTS and WLANspecifications were based on rather different set of requirements and they ended up supporting ratherdifferent set of QoS features Consequently, the QoS consistency turns out to be a quite challengingissue To provide more insight on this issue, we discuss below a list of WLAN QoS deficiencies withrespect to UMTS QoS Apparently, when we target multimedia session continuity across UMTS andWLAN, we should carefully take these deficiencies into consideration and understand their impact Thediscussion is based on the assumption that the WLAN MAC layer complies with IEEE 802.11 [26] plusthe amendments of IEEE 802.11e [17] and the physical layer complies with IEEE 802.11g [18] For agood introduction to the QoS aspects of IEEE 802.11e the reader is referred to Chapter 5 as well as to[28]

(1) In a WLAN we cannot support unequal error protection across different media streams Forinstance, an AMR payload, an H.263 payload and a HTTP payload will all be subject to the samechannel coding (for a given transmission rate) and therefore all media streams will be equallyprotected against transmission errors, no matter what their different bit error rate requirements are

On the contrary, in UTRAN different radio transport channels (each with their own channel coding)can be established for streams with different error rate requirements Thus, error rate can becontrolled on a per stream basis

(2) Unequal error protection in a WLAN cannot also be supported between different flows in the samemedia stream For instance, the class A and class B bits of an AMR payload [29] will be equallyprotected against transmission errors, although class B bits can tolerate a higher bit error rate In

fact, the WLAN layers are unable to distinguish the different flows of the AMR stream On thecontrary, UTRAN typically employs different radio transport channels for the different AMR flows.(3) Although different media streams may tolerate different residual bit error rates, in a WLAN there is

no way to control the residual bit error rate This is because the WLAN has been optimized tosupport data streams and therefore enforces a very small residual bit error rate (by using 32-bit longCRC codes) In practice, this may result in increased MAC Service Data Unit (MSDU) loss rate(especially when acknowledges are not used) since all erroneous packets will be dropped even ifsome of them could be tolerated by the application Moreover, the channel utilization will bedecreased because, according to 802.11 [26], when a station receives an erroneous MSDU it cannotaccess the channel until after an Extended Inter-Frame Space (EIFS) period

(4) WLAN stations cannot have dedicated radio channels as in UTRAN and therefore the queuingdelay and jitter figures could be increased Note that in 802.11e Hybrid Coordinator ChannelAccess (HCCA) mode the stations transmit with a polling discipline and hence delay and jitter willdepend on the overall number of stations requesting resources and on the scheduler characteristics

In 802.11e Enhanced Distributed Channel Access (EDCA) mode the stations transmit with arandom access discipline tailored to support several different traffic classes

(5) In the WLAN there is no adaptive mechanism for controlling the MSDU loss rate in real-time Thetypical way to control MSDU loss rate is via link adaptation with transmission rate change.However, this adaptation is not mandatory in all WLAN stations, it is implementation dependentand, more importantly, it is typically based on some predefined loss rate thresholds that do notcorrelate directly with the loss rate requirements of the transmitted media streams The IEEE802.11g standard allows the transmit power level to vary (the same holds true for 802.11b and802.11a) but in practice all WLAN stations tend to use the maximum power level at all times, since

no fast power control mechanism exists Note also that in 802.11e EDCA loss rate is even harder tocontrol since collisions are unavoidable

(6) There are no soft handovers in WLAN Handovers are typically hard in nature, i.e., follow thebreak-then-make approach, and hence considerable transmission disruptions may exist that result inQoS degradation Moreover, handovers in 802.11 are solely controlled by the WLAN stations, sothe WLAN infrastructure cannot provide tight control of the QoS provisioning If a WLAN stationtends to ‘ping-pong’ between two APs the QoS will be severely affected and the WLANinfrastructure has no means to prevent that

One of our main conclusions is that the ‘Service Data Unit (SDU) Error Ratio’ and the ‘Residual BitError Ratio’ attributes used in UMTS QoS profile (see [30]) cannot be negotiated and controlled in an802.11 WLAN, mainly due to physical layer restrictions Also, the WLAN infrastructure is nearlyimpossible to guarantee a strict QoS levelegiven that there is no standardized mechanism for softhandovers Of course, these deficiencies represent the price we pay for facilitating simple and cost-efficient WLAN designs

Nevertheless, it is important to stress that the above QoS deficiencies of WLANs do not necessarilymean that seamless session continuity from UMTS to WLAN cannot be supported Under certainconditions, seamless session continuity can be provided (but not guaranteed) even with inefficientutilization of WLAN radio resources (but these are cheap anyway!) To validate our point, in the nextsessions we carry out a performance study and evaluate the number of UMTS roamers that can beadmitted to the WLAN under certain restrictions, e.g., maintain the QoS level that was experienced inUMTS, respect WLAN policies, etc

e It is interesting to observe that IEEE 802.11e carefully refers to the provided QoS as ‘differentiated’ and

‘parameterized’ QoS, not as ‘guaranteed’ QoS.

7.7 Performance Evaluation

Given the aforementioned QoS discrepancies across UMTS and WLAN and, in particular, the limitedQoS features of WLAN with respect to UMTS, it is interesting to investigate the feasibility andconstraints of seamless QoS provision In this context, we are interested in investigating the capability ofWLAN to support seamless QoS provision for multimedia sessions that have previously been initiated inthe UMTS environment Say, for example, that the WLAN starts accepting UMTS roamers, each onewith a video session in progress Will these video sessions experience a QoS level consistent with theQoS level provided before in the UMTS environment? Under what conditions is this possible? And howmany such video sessions can be admitted into the WLAN without compromising the requirements forseamless QoS provision and comply with the possible interworking policy of the WLAN? These are theimportant questions that we are dealing with in this section

To derive realistic answers to the above questions we consider the interworking scenario illustrated inFigure 7.13 In this scenario, there is only one AP, which provides access services to two classes ofusers: (a) the WLAN data users and (b) the UMTS roamers, who are UMTS subscribers handed overfrom UMTS All UMTS roamers are considered statistically identical and the same holds true for theWLAN data users In addition, the AP as well as the UMTS roamers comply with the procedures andelements specified in IEEE 802.11i (or Wi-Fi Protected Access) for enhanced security provision This isrequired for supporting UMTS-based access control and having the UMTS roamers authenticated andauthorized by their UMTS home environment, as specified in [22, 23] All WLAN stations, includingthe AP, support the extended MAC layer specified in IEEE 802.11e and the physical layer specified inIEEE 802.11g We assume that the transmission rate of WLAN stations is 24 Mbps and that notransmission errors occur (the channel is ideal)

Each WLAN data user has a number of ongoing non-real-time data sessions (e.g., web browsing,e-mail, ftp, etc.), which generate an aggregate user traffic described as a Poisson process with 256 kbpsmean rate On the other hand, each UMTS roamer has a uni-directional (uplink) real-time video session

in progress, which has been initiated in the UMTS domain and granted the negotiated QoS parameters

Figure 7.13 The interworking scenario considered for performance evaluation.

shown in Table 7.2 The selection of these parameters is based on [31] and the assumption that H.263video coding is used with a target bit rate of 64 kbps In our simulations the video packet traffic isgenerated with the aid of the video trace files found in http://trace.eas.asu.edu/TRACE/ltvt.html and theuse of Real Time Protocol (RTP) and Real Time Control Protocol (RTCP).

For the interworking scenario discussed above, our main goal is to evaluate how many UMTSroamers the WLAN network can support under the following two constraints

(1) The video streams of all UMTS roamers admitted to the WLAN must experience at least the sameQoS level as the one negotiated in the UMTS network (see Table 7.2) For example, the MAC SDU(MSDU) loss rate in the WLAN must not exceed the corresponding UMTS SDU error ratio, i.e.,

103 This constraint is required for satisfying the seamless service continuity requirements.(2) The bandwidth available to WLAN data users must not diminish below a predefined threshold Thisconstraint makes it possible to enforce a bandwidth reservation policy consistent with the WLANoperator’s interworking requirements For example, the WLAN operator may need to ensure thatthe WLAN data users will have at least 7 Mbps of bandwidth available no matter how many UMTSroamers are admitted into the WLAN Therefore, UMTS roamers can be served with possiblyhigher priority than WLAN data users (in order to meet the seamless continuity requirements) butthey cannot consume all the available bandwidth For this purpose, the WLAN needs to apply anadmission control function that would reject further association requests from UMTS roamersshould the bandwidth reservation limit be reached In our study we assume that such an admissioncontrol function is implemented and we take it into account for calculating the maximum number

of UMTS roamers that can be admitted into the WLAN

7.8 Performance Results

For performing our evaluations we consider two practical WLAN deployment scenarios: (i) aContention-based Scenario, where the WLAN AP does not support HCCA access mode and thus

Table 7.2 The QoS values negotiated in UMTS for the H.263 video sessions

UMTS QoS parameter Parameter value Comments

Traffic class Conversational

Residual BER 105 Corresponds to 16-bit CRC This cannot be controlled

in the WLAN since the supported CRC is always

32 bits long.

Maximum SDU size 1500 bytes

Guaranteed bit rate for uplink 64 kbps

Maximum bit rate for uplink 128 kbps

Guaranteed bit rate for downlink 2 kbps To support RTCP traffic.

Maximum bit rate for downlink 2 kbps

Delivery order No Re-ordering should be taken care by application in

order to minimize the transfer delay.

Delivery of erroneous SDUs No

Traffic handling priority Subscribed Not relevant

SDU format information Not used Unequal error protection within the bits of the same

SDU is not required.

Allocation/retention priority Subscribed Not relevant

Source statistics descriptor Unknown Only used for voice and allows the network to assess

the utilization of trunk resources.

both UMTS roamers and WLAN data users employ contention-based channel access, and (ii) aContention-Free Scenario, where the WLAN AP supports HCCA access mode and all UMTS roamersare serviced in this mode (i.e., contention free).

7.8.1 Contention-Based Scenario

In this scenario we assume that both UMTS roamers and WLAN data users use contention-basedchannel access A key policy applied by the WLAN indicates that at least L Mbps must be available tothe WLAN data users Hence, UMTS roamers can be admitted to the system as long as (i) the WLANcan support L Mbps of data traffic and (ii) the QoS experienced by the video streams meets or exceedsthe QoS negotiated in the UMTS environment (according to Table 7.1)

First, we consider a typical case that is expected during the early deployment of UMTS/WLANinterworking, where the WLAN data users use legacy terminals with no 802.11e support These usersaccess the channel by employing the Distributed Coordination Function (DCF) with the followingaccess parameters (see [26]): DCF Inter-Frame Space (DIFS)¼ 2 slots, minimum Contention Window(CWmin)¼ 15, maximum Contention Window (CWmax) ¼ 1023 and Persistence Factor (PF) ¼ 2 Onthe contrary, UMTS roamers use 802.11e aware terminals and employ an EDCA access class with thefollowing access parameters (see [17] and Chapter 5): Arbitration Inter-Frame Space Number(AIFSN)¼ 3, CWmin ¼ 15, CWmax ¼ 1023 and PF ¼ 2 All terminals maintain uplink buffers thatcan hold up to eight maximum sized MSDUs The terminals of UMTS roamers make every effort totransmit all video packets within their delay bound, which is considered equal to 40 ms for consistencywith UMTS If a video packet, however, is delayed for more than 40 ms, it is dropped This policyguarantees that the delay experienced by all successfully transmitted video packets will be less than

40 ms Hence, the key parameter affecting the QoS of video streams will be the loss rate, which should

be kept below the corresponding UMTS limit (103)

Our simulation results for the above case reveal that the limiting factor for the maximum number ofUMTS roamers in the WLAN is not the bandwidth reservation constraints but rather the loss rate ofvideo streams As illustrated in Figure 7.14, the MSDU loss rate for video traffic reaches the UMTS

0.0E+00 5.0E-04 1.0E-03 1.5E-03 2.0E-03 2.5E-03 3.0E-03

4 8 12 16 20 24 28 32 36 40 44 48 52 56 60 64 68

Number of UMTS roamers

negotiated value (103) when there are 56 UMTS roamers (or equivalently 56 video streams) for

L¼ 5 Mbps, or 24 UMTS roamers for L ¼ 7 Mbps, or eight UMTS roamers for L ¼ 8 Mbps.Apparently, when L increases (i.e., when there are more WLAN data users in the system), thetransmission delay of video packets is increased and the probability to reach the 40 msec delaybound is increased as well Hence, the video packet loss rate rises accordingly

In Figure 7.15 we display the maximum delay of 99% of the successfully delivered packets for bothUMTS roamers and WLAN data users As expected, the delay of video packets is always larger than thedelay of WLAN data traffic, since the latter employs a smaller inter-frame space and thus gains priorityover video traffic We also note that the maximum delay experienced by the video packets in the WLANdomain is far less than the limit of 40 msec that was negotiated in the UMTS domain For example, asshown in Figure 7.15, when L¼ 7 Mbps and there are 24 UMTS roamers (thus the loss rate is within theUMTS negotiated limit) the maximum delay is about 15 msec This leads to a significant conclusion:The WLAN can meet or exceed the QoS negotiated in the UMTS domain but in a very inefficient way.Indeed, we can readily derive that, when the transmission rate is 24 Mbps and the WLAN data usersoffer 7 Mbps aggregate data traffic, there will be about 4.3 Mbps available for UMTS roamers (thiscorresponds to 47% max channel utilization) However, only up to 1.5 Mbps (24 64 kbps) can beutilized for meeting the loss rate requirements

We consider another case where both UMTS roamers and WLAN data users are 802.11e aware Thetwo user classes are mapped however to different EDCA access classes with the UMTS roamers givenhigher access priority for meeting their demanding QoS needs In particular, we assume that the UMTSroamers are mapped to an EDCA access class with the following access parameters: AIFSN¼ 3,CWmin¼ 6, CWmax ¼ 511 and PF ¼ 2 Similarly, the WLAN data users are mapped to an EDCAaccess class with the following access parameters: AIFSN¼ 6, CWmin ¼ 31, CWmax ¼ 1023 and

PF¼ 2

In contrast to the previous case (legacy WLAN data users), our simulation now indicates that the lossrate experienced by video packets is almost negligible since the UMTS roamers are given preferential

110100

Figure 7.15 The max delay of 99% of the delivered MSDUs for both the video and the data traffic vs the number of UMTS roamers admitted in the WLAN.

access to the wireless medium Therefore, the limiting factor for the maximum number of UMTSroamers in the WLAN is not the loss rate of video streams but rather the bandwidth reservationconstraints Indeed, as displayed in Figure 7.16, when L¼ 7 Mbps, the MSDU loss rate for data traffic isequal to zero for up to 40 UMTS roamers Up to this number of roamers, the capacity offered to theWLAN data users is indeed 7 Mbps and hence the bandwidth reservation policy is respected However,when more than 40 UMTS roamers are admitted to the WLAN, this policy cannot be satisfied as thebandwidth utilized by WLAN data users is quickly diminished.

From the above, it can easily be deduced that we can experience significant capacity benefits when theWLAN data users become 802.11e aware and the appropriate access parameters are used; e.g., from 24UMTS roamers with legacy WLAN data users we can climb to 40 UMTS roamers with 802.11e awareWLAN data users Similar observations can be made in the case of the other two scenarios, namelywhen L¼ 5 and L ¼ 8, where the maximum number of UMTS users that can be supported increases to

84 and 12 respectively Yet, this gain is achieved with a cost on the delay performance of WLAN datausers For example, as illustrated in Figure 7.17, when L¼ 7 the maximum delay experienced by theWLAN data users, when there are 40 UMTS roamers in the system is about 140 msec In the case oflegacy WLAN data users, however, the delay is about 10 msec when there are 24 UMTS roamers (seeFigure 7.15) By comparing Figures 7.15 and 7.17, it becomes evident that when the WLAN data usersbecome 802.11e aware, they can still use the same total bandwidth (7 Mbps) but they experienceconsiderably increased delay, which accounts for the considerably more UMTS roamers in the WLAN.Moreover, not only can we have more UMTS roamers admitted in the WLAN but each one willexperience reduced delay, i.e., 2–4 msec in Figure 7.17 as compared with 7–15 msec in Figure 7.15

7.8.2 Contention-Free Scenario

In the contention-free scenario it is assumed that all UMTS roamers operate in HCCA mode andtherefore they do not contend with the WLAN data users HCCA implements a polling mechanismwhich allows the Hybrid Coordinator (HC) entity of 802.11e (normally implemented in the AP) to

control the access to the wireless channel by assigning Transmission Opportunities (TXOPs) to therequesting WLAN terminals Since the access to the channel is centrally controlled and there is nocontention or collisions, HCCA is appropriate for providing parameterised QoS services with specificbounds In our simulation, each handover of an H.263 video session from UMTS triggers theestablishment of a new 802.11e Traffic Stream (TS) with the signaling messages illustrated at the end

of Figure 7.12 The traffic characteristics and QoS requirements of each TS are described by the TrafficSpecification (TSPEC) element shown in Table 7.3 (see Chapter 5 and [17] for details on theseparameters)

1 10 100 1000

Figure 7.17 The maximum delay of 99% of the delivered MSDUs for both the video and the data traffic vs the number of UMTS roamers admitted in the WLAN.

Table 7.3 The TSPEC values negotiated in WLAN for the H.263 video sessions

802.11e TSPEC parameter Parameter value Comments

Nominal MSDU Size 320 bytes Calculated from the H.263 trace file

Maximum MSDU Size 1917 bytes Calculated from the H.263 trace file

Minimum service interval 0 msec ‘Don’t care’

Maximum service interval 40 msec Considered equal to delay bound

Mean data rate 64 kbps Calculated from the H.263 trace file

Peak data rate 383.4 kbps Calculated from the H.263 trace file

Maximum burst size 1917 bytes Calculated from the H.263 trace file

Delay bound 40 msec For consistency with UMTS (see Table 7.1) Minimum PHY rate 24 Mbps All stations can transmit with 24 Mbps

The allocation of TXOPs to UMTS roamers is performed by the Scheduler (implemented by the HC)and is based on the QoS requirements defined by the corresponding TSPEC The scheduler decides forboth the polling time of a UMTS roamer as well as the duration of the allocated TXOP For the purposes

of our simulation, it is assumed that the WLAN implements the reference scheduler described in [17],which is also referred to as the Simple Scheduler Based on the negotiated Mean Data Rate and DelayBound of each video session, the Simple Scheduler calculates a fixed TXOP length for each UMTSroamer and a Service Interval (SI) The result is that the scheduler implements a periodic service pattern

by sequentially serving all UMTS roamers every SI time units

The impact of the WLAN data users, which still operate in contention mode, is considered by limitingthe percentage of channel time available to the Simple Scheduler In particular, assuming again the samebandwidth reservation policy, i.e., allowing 7 Mbps minimum bandwidth for the WLAN data users, wereadily calculate that about 40% of channel time is left for HCCA operation Given that constraint, ourmain objective is to assess the performance of the Simple Scheduler in terms of channel utilization andmaximum number of UMTS roamers that can be supported

Figure 7.18 depicts the Channel Occupancy (i.e., the fraction of the total channel time spent in HCCAmode) versus the UMTS roamers admitted in the WLAN Observe that the occupancy increases linearlyand that the 40% of channel capacity available for HCCA is reached for 17 UMTS roamers Thisexposes a fairly inefficient channel utilization Indeed, with 40% of the available radio resources up toapproximately 4.8 Mbps can be accommodated, yet the Simple Scheduler manages to accommodateonly 1.06 Mbps (17 64 kbps)

In contrast to the contention-based scenario, video packets transmitted in HCCA mode are not lostbecause the scheduler tries to respect the negotiated delay bounds and allocate enough TXOPs toaccommodate all traffic load offered by the UMTS roamers Therefore, the QoS experienced by theUMTS roamers in HCCA mode is affected only by the delay characteristics Figure 7.19 illustrates thedelay experienced by the delivered video packets, which is nearly constant throughout the consideredrange of UMTS roamers and remains below the delay bound negotiated in the UMTS domain (40 msec)

It is interesting to note, however, that this delay is larger than the corresponding delay in based scenario Comparing for example Figure 7.19 with Figure 7.17, we identify that the delay incontention-free scenario increases by nearly four times This is a result of the inefficient TXOPallocation of the Simple Scheduler

contention-As already pointed out, the main issue in HCCA operation is the inefficient resources utilization andhence the support of a relatively small number of users (17) as compared with the contention-based

0 5 10 15 20 25 30 35 40 45 50

scenario However this should not be attributed to the HCCA mechanism itself but rather to the inherentcharacteristics of the Simple Scheduler, which employs a greedy, over-provisioning strategy for TXOPallocation in order to meet the QoS requirements Our simulation results denote that the TXOP LossFactor, i.e., the percentage of unused time allocated to a UMTS roamer (because there were no videopackets for transmission), is a bit more than 80% This basically indicates that the Simple Schedulerallocates four times more radio resources to a UMTS roamer than what is required This happensbecause when the Simple Scheduler calculates the TXOP duration per UMTS roamer, it makes a worst-case estimation so as to be able to accommodate the largest MSDU size With Constant Bit Rate (CBR)traffic, where the MSDU size features nearly no variation, this sounds like a reasonably simple andacceptable approach However, with Variable Bit Rate (VBR) traffic like the considered H.263 videostreams, this strategy leads to excessive inefficiency and thus the performance of the contention-freescenario does not compare favorably with the performance of the contention-based scenario To improvethe scheduling efficiency for VBR traffic, more sophisticated schedulers have been proposed in thetechnical literature; these adapt the TXOP lengths to the actual size of buffered MSDUs Examples ofsuch schedulers are presented in [15,32].

We moved on to emphasize the importance of traffic scheduling in WLANs, especially in IEEE802.11e networks To prove our concept, we described a proposed scheduling algorithm and compared itwith a simple solution included in the IEEE 802.11e specification An interesting challenge for futureresearch in the area of QoS over WLANs is the combination of different techniques and mechanismsthat can support both efficient transmission over the wireless medium, and fast but yet simple handoverexecution, with strict QoS guarantees for different kinds of traffic This will allow for the use of a wide

0 2 4 6 8 10

range of multimedia applications over WLANs and extend the set of environments that suchtechnologies can operate.

Another major problem in providing QoS for wireless multimedia is the low performance of RSVP inwireless networks, as a result of its design for fixed environments with stable links Towards solving thisproblem, we presented one of the most important proposals, namely, the dynamic RSVP (dRSVP).dRSVP extends the standard RSVP protocol to use ranges of traffic specifications, instead of absolutevalues The benefit is that intermediate nodes can adjust allocations, as long as these are within thespecified ranges, depending on changes of the available bandwidth in the output links After presentingthe basic features of dRSVP, we focused on a component that is essential for its performance, i.e., theflow rejection algorithm The presented algorithm was found to considerably improve the protocol’sperformance, especially in environments with heavy bandwidth fluctuations The main characteristic ofthis algorithm is that it rejects the minimum number of flows required to reduce the overall bandwidthrequirements below the offered capacity at any time As for the future, research at international level ismoving towards designing new QoS provision protocols, assuming both fixed and wireless environ-ments This means that the resulting protocols should be able to handle both user mobility and unstablelinks

Finally, we focused on the problem of WLANs interworking with UMTS Undoubtedly, there are stillseveral challenges to be addressed for enabling the seamless interworking of wireless LAN and UMTSnetworks As we discussed in this chapter, the QoS discrepancies between these networks raise some ofthese challenges Although the WLAN QoS capabilities have recently been extended considerably withthe introduction of IEEE 802.11e, the WLAN is still incapable of supporting all the QoS featuresprovided by UMTS (e.g., to dynamically control the MSDU error rate, the residual bit error ratio,unequal error protection, etc.) This is partially attributed to the characteristics of the WLAN physicallayer, which has been kept relatively simple in order to enable low-cost designs As a result, multimediatransmission over WLANs turns out to be less efficient than the UMTS (but definitely less expensive).Vertical handovers also present another challenge for seamless UMTS/WLAN interworking Since theyare usually implemented as mobile-controlled, hard handovers, they bring up considerable QoSconcerns, which severely affect the provision of seamless interworking

In our study of various interworking scenarios, where UMTS subscribers with ongoing real-timevideo sessions handed over to WLAN, we mainly quantified the maximum number of UMTSsubscribers that can be admitted to the WLAN, subject to maintaining the same level of UMTS QoSand respecting the WLAN bandwidth reservation policies Our results suggest that the WLAN cansupport seamless continuity of video sessions for only a limited number of UMTS subscribers,depending on the bandwidth reservations, on the WLAN access parameters and on the QoS require-ments of video sessions The operation of the Simple Scheduler was proved inefficient for video trafficand consequently the contention-based scenario gave a better performance, even though the video traffichad to content with data traffic for channel access