Performance Evaluation of the AAL2 protocol within the UTRAN

Transcription

1 1809-Her/Telecom 58/7-8 4/08/03 13:31 Page 1 pp Performance Evaluation of the AAL2 protocol within the UTRAN Rani MAKKÉ*, Samir TOHMÉ*, Jean-Yves COCHENNEC**, Sophie PAUTONNIER*** Abstract The AAL2 protocol was chosen as a transport protocol within the UMTS Terrestrial Radio Access Network (UTRAN). This protocol handles the radio channels between the mobile terminal and the Radio Network controller (RNC). On radio channels, the time constraints are stringent because of the synchronisation mechanisms at the MAC layer. The AAL2 protocol should guarantee a certain Quality of Service (QoS) for the traffic handled on the AAL2 connections. In this paper, we introduce a general description of the AAL2 protocol and its position in UTRAN and we study some performance issues of this protocol. We propose two schemes to transport the AAL2 traffic on the Iub and Iur interfaces and we study different scheduling algorithms (FCFS, RR, WRR, EDF, Priority) at AAL2 level in order to evaluate the performance of each algorithm. The optimal Timer-CU value is studied in this paper in order to choose a trade-off between time constraints and bandwidth efficiency. The equivalent bandwidth is calculated in order to evaluate the capacity of a VC supporting AAL2 connections. Finally, the AAL2 switching technology is compared with the ATM switching technology in the case of a traffic concentrator in order to evaluate the advantages of the AAL2 switching. Keys words: Résumé ÉVALUATION DES PERFORMANCES DU PROTOCOLE AAL2 DANS L UTRAN Le protocole AAL2 a été choisi comme protocole de transport dans le réseau terrestre d accès radio de l UMTS (UTRAN). Ce protocole transporte les canaux radio entre le terminal mobile et le RNC (Radio Resource Controller). Sur les canaux radio, les contraintes de temps sont strictes à cause des mécanismes de synchronisation de la couche MAC. Le protocole AAL2 doit garantir une certaine qualité de service pour le trafic transporté sur les connexions AAL2. * GET/Télécom Paris, 46 rue Barrault Paris CEDEX 13 France Rani.Makke@enst.fr ** France Télécom R&D, 2 avenue Pierre Marzin Lannion, France jeanyves.cochennec@rd.francetelecom.fr ***Sophie Pautonnier, Mitsubishi Electric ITE-TCL, 1 allée de Beaulieu, CS 10806, Rennes CEDEX 7, France pautonnier@tcl.ite.mee.com 1/25 ANN. TÉLÉCOMMUN., 58, n 7-8, 2003

2 1809-Her/Telecom 58/7-8 4/08/03 13:31 Page 2 2 R. MAKKÉ PERFORMANCE EVALUATION OF THE AAL2 PROTOCOL WITHIN THE UTRAN Dans cet article, nous décrivons le protocole AAL2 et sa position au sein de l UTRAN et nous étudions quelque aspects de performance de ce protocole. Nous proposons deux schémas pour transporter le trafic AAL2 sur les interfaces Iub et Iur et nous étudions différents algorithmes d ordonnancement (FIFO, RR, WRR, EDF, Priorité) au niveau AAL2 pour évaluer les performances de chaque algorithme. La valeur optimale du Timer-CU est aussi étudiée pour choisir un compromis entre les contraintes de délai et l utilisation de la bande passante. La bande passante équivalente est calculée pour évaluer la capacité d un VC transportant des canaux AAL2. Enfin, La commutation AAL2 est comparée avec la commutation ATM dans le cas de concentration de trafic pour évaluer les avantages de la commutation AAL2. Mots clés : Contents I. Introduction II. The AAL2 protocol III. The UTRAN architecture IV. AAL2 in UTRAN V. Performance issues VI. Traffic model VII. Simulation model VIII. Simulation results IC. Conclusion References (18 réf.) I. INTRODUCTION Asynchronous Transfer Mode (ATM) is now widely used in many core networks to transport high data rate with a highly reliable Quality of Service (QoS) in particular for multimedia applications. In 1997, ITU-T has standardised ATM Adaptation Layer 2 (AAL2) which allows multiplexing of several small packets called mini-cells in one ATM cell [1,2]. This is more and more useful with the introduction of new low bit-rate codec. AAL2 is therefore well suited for transport in the Core Network, in particular for real-time applications. Another advantage of the AAL2 protocol is its switching capability as each AAL2 mini-cell contains a header including an AAL2 channel identifier. In IMT2000/UMTS, AAL2 was chosen as a transport technology for the radio access network UTRAN (UMTS Terrestrial Radio Access Network) in release 99 of the 3GPP standard [3]. The transmission delays in UTRAN have to be minimised as much as possible because of the real time services carried over UTRAN. In fact, even non real-time traffic coming from the core network is becoming more or less real-time inside UTRAN because of the air interface synchronisation. As the delay requirements are very important in the access network, a precise evaluation of the performance of the AAL2 protocol is needed. This paper presents the main theoretical results of a collaborative project (RNRT MINICEL Project) between the ENST (Télécom-Paris), France Télécom R&D and Mitsubishi Electric ITE in which a prototype including a UTRAN environment simulator and an AAL2 switch has been developed [4]. This work was supported by the French Ministry of Industry. ANN. TÉLÉCOMMUN., 58, n 7-8, /25

3 1809-Her/Telecom 58/7-8 4/08/03 13:31 Page 3 R. MAKKÉ PERFORMANCE EVALUATION OF THE AAL2 PROTOCOL WITHIN THE UTRAN 3 II. THE AAL2 PROTOCOL The AAL2 protocol has been standardised to transport very low bit-rate applications with real-time constraint and variable bit-rate (e.g. the compressed voice). AAL2 was defined to get around the problem of the ATM cell packetization delay that becomes critical for the low bit-rates (at 16kbps, its value is 24 ms). The solution is simple : when multiplexing several communication flows in the same ATM channel, the delay becomes reasonable for a given communication. The AAL2 protocol consists of variable length data units called mini-cells with a maximum payload length of 45 bytes (optionally 64 bytes). The AAL2 layer is divided into two sub-layers : the SSCS (Service Specific Convergence Sub-layer) [5] and the CPS (Common Part Sub-layer). The SSCS is divided into three sublayers : the SS-ADT (Service Specific Assured Data Transfer), the SS-TED (Service Specific Transmission Error Detection) and the SS-SAR (Service Specific Segmentation And Reassembly) (Fig. 1). Only SS-SAR is used in UTRAN. It segments higher-level data units exceeding 45 bytes (optionally 64 bytes) into packets with a maximum length of 45 bytes (optionally 64 bytes). The CPS layer overloads each mini-cell by a 3-byte header, whose format is described in FIG. 1 AAL2 sub-layers. Sous-couches AAL2. 3/25 ANN. TÉLÉCOMMUN., 58, n 7-8, 2003

4 1809-Her/Telecom 58/7-8 4/08/03 13:31 Page 4 4 R. MAKKÉ PERFORMANCE EVALUATION OF THE AAL2 PROTOCOL WITHIN THE UTRAN Figure 2. The CID (Channel IDentifier) field identifies the AAL2 connection. There are 256 possible CID values, 8 of them are reserved for signalling purpose, and the rest may be used to identify 248 different AAL2 connections. The LI (Length Indicator) field determines the minicell payload length. By default, the maximum length is 45 bytes, but it may be 64 bytes if there is an indication at the connection establishment procedure. The UUI field is assigned to the SSCS. Its 5 bits (32 code-points) are not interpreted by the CPS sub-layer and they are passed transparently from the SSCS transmitting entity to the SSCS receiving entity. The HEC (Header Error Control) is used for error detection in the mini-cell header. CID LI UUI HEC 8 bits 6 bits 5 bits 5 bits FIG. 2 AAL2 minicell header format. Format de l en-tête d une mini-cellule AAL2. The ATM header allows two levels of addressing (Virtual Path Identifier VPI and Virtual Circuit Identifier VCI). Thus, it is possible to set up ATM VPCs between AAL2 end-points and to allow them to use VCI and CID to create multiple native connections. With a 16-bit VCI field, an ATM VPC will be able to support up to AAL2 connections. The mini-cells are inserted in the ATM cells and 1-byte field (STF : STart Field) is added at the beginning of the ATM payload. The STF contains a pointer to the first byte of the first minicell header in the ATM payload. Overlapping is used : one minicell inserted in the ATM cell can overlap onto the next cell (Fig. 3). Padding may be added at the end of the ATM cell payload if there are no additional minicells to be inserted. FIG. 3 Overlapping Chevauchement. ANN. TÉLÉCOMMUN., 58, n 7-8, /25

5 1809-Her/Telecom 58/7-8 4/08/03 13:31 Page 5 R. MAKKÉ PERFORMANCE EVALUATION OF THE AAL2 PROTOCOL WITHIN THE UTRAN 5 The Timer-CU is an important parameter of the AAL2 protocol. When a CPS-SDU (CPS Service Data Unit) arrives in the CPS sub-layer, the CPS protocol adds the 3-byte header to form the CPS-packet. Then the Timer-CU is armed and STF is calculated. CPS-PDU is then transmitted to the ATM layer either when it is full, i.e. the 47-byte payload field is filled, or when the Timer-CU has expired. If the Timer-CU expires before the CPS-PDU becomes full, padding is added to fill CPS-PDU that is transmitted to the ATM layer. III. THE UTRAN ARCHITECTURE The UTRAN is the UMTS Terrestrial Radio Access Network [6]. Its architecture is very similar to the architecture of the GSM radio access network. Figure 4 represents the general architecture of UTRAN. The different elements of UTRAN are: RNC - Radio Network Controller : it controls the radio resources. Node B the equivalent of the BTS (Base Transceiver Station) in GSM. The principal role of this node is to transmit data and signalling on the radio interface. RNS - Radio Network Subsystem : it is the access part of the UMTS network that manages the allocation and the release of the radio resources for a set of cells. There is only one RNC in each RNS. The different UTRAN interfaces are: Iu : interface between the RNS and the Core Network [7]. Iub : interface between one Node B and one RNC [8,9]. Iur : interface between two RNCs [9,10]. When the mobile terminal is in soft handover state (when the terminal has several radio links with different cells corresponding to different RNS), one RNS has an inter-connection point with the Core Network. This RNS is called S-RNS (Serving RNS). The other RNS is called D-RNS (Drift RNS) and can transmit all the user data flows to the S-RNC which establishes the recombination. The recombination principle consists of recombining all the data flows received from one user into one data flow to the core network. IV. AAL2 IN UTRAN The AAL2 protocol is deployed on the Iub and Iur interfaces within UTRAN for real-time and non real-time flows. On these interfaces, the AAL2 protocol supports radio channels [11] extended between the UE (User Equipment) and the RNC as shown in Figure 5. The RLC (Radio Link Control) layer [12] is transparent for voice flows. For data flows, the RLC layer segments data packets received from the core network into several RLC-PDUs (RLC - Packet Data Unit) with a maximum size predefined by the RRC (Radio Resource Control) functions [13]. 5/25 ANN. TÉLÉCOMMUN., 58, n 7-8, 2003

6 1809-Her/Telecom 58/7-8 4/08/03 13:31 Page 6 6 R. MAKKÉ PERFORMANCE EVALUATION OF THE AAL2 PROTOCOL WITHIN THE UTRAN FIG. 4 UTRAN general architecture. Architecture générale de l UTRAN. The MAC (Medium Access Control) [14] layer selects a number of Transport Blocks (TB) to be sent in one TTI (Transmission Time Interval). Each TB contains one RLC-PDU and the number of TBs sent in one TTI is predefined by the RRC functions. TTI is the duration of a radio frame and it is a multiple of 10 ms. It may be typically 10, 20, 40 or 80 ms. At this time, all MAC-PDUs are given a CFN (Connection Frame Number) which is a kind of timestamp indicating on which radio frame this data should be sent. This is why, as already mentioned above, all flows (voice and data) get time constraints on the UTRAN interfaces even though a tolerant margin may be chosen for non real-time traffic. If this data arrives too late in the Node B (e.g. case of downlink), the frame is discarded. The FP (Frame Protocol) layer assembles in the same frame the different TBs sent in the same TTI and adds some more information which allows the receiving entity to know how the frame is built (number and size of TBs). In the uplink, some information about the quality of ANN. TÉLÉCOMMUN., 58, n 7-8, /25

7 1809-Her/Telecom 58/7-8 4/08/03 13:31 Page 7 R. MAKKÉ PERFORMANCE EVALUATION OF THE AAL2 PROTOCOL WITHIN THE UTRAN 7 FIG. 5 Iub protocol stack. Pile protocolaire de l interface Iub. the received frame is added, so that the S-RNC has an idea of the reliability of the data from one link compared to another one. The AAL2 layer segments FP-PDUs (FP - Packet Data Unit) when their size exceed 45 bytes (optionally 64 bytes) in the SS-SAR sub-layer and then inserts the CPS packets (the mini-cells) in the ATM cells. V. PERFORMANCE ISSUES Within UTRAN, two major categories of flows are transported on the Iub and Iur interfaces : real-time applications (e.g. voice) which require low delay and non real-time applications (e.g. web browsing) which are tolerant of transfer delay. Even if all flows become delay sensitive in UTRAN because of the air interface synchronization, non real-time applications may tolerate a higher delay than real-time applications. In the case of mixed traffic (real-time and non real-time flows), we can privilege real-time packets in order to meet stringent delay requirements [15]. Non real-time packets are delivered with a higher delay but without violating the time constraints of the air interface. Therefore, differentiation between services is needed at the AAL2 layer. To transport these two types of flows on AAL2 connections, two schemes are foreseen: 1. Mono-service VC scheme : in this scheme, real-time flows are aggregated in one ATM VC with stringent class of QoS and treated separately from non real-time applications which are aggregated in another ATM VC with tolerant class of QoS. In this scenario, real-time VC transports one type of traffic and it is treated at ATM layer with higher priority than the non realtime VC. 2. Multi-service VC : in this scheme, real-time and non real-time applications are aggregated in the same VC. In order to differentiate between services, a scheduling mechanism is 7/25 ANN. TÉLÉCOMMUN., 58, n 7-8, 2003

8 1809-Her/Telecom 58/7-8 4/08/03 13:31 Page 8 8 R. MAKKÉ PERFORMANCE EVALUATION OF THE AAL2 PROTOCOL WITHIN THE UTRAN needed at the AAL2 layer to manage priority between different queues. At the ATM layer, one type of VC is needed with a stringent class of QoS. In the first scheme, all connections in the same VC have the same class of service. Thus, they have the same priority and a simple fair scheduling algorithm is recommended. We propose the classic FCFS (First Come First Served) and the RR (Round Robin) algorithms. The FCFS or FIFO (First In First Served) consists of providing one queue for all the packets coming from all the AAL2 connections in the VC. Since all the packets have the same time requirements because the traffic is homogeneous, the first packet arriving in the queue must be served first and the FCFS algorithm is suitable. The RR algorithm consists of providing one queue for each AAL2 connection, and the scheduler selects one packet from each queue in each cycle. In the second scheme, different classes of service are handled in the same VC. An appropriate scheduling algorithm should be implemented in the CPS multiplexer which selects the packet that should be sent on the link. The FCFS policy is not an appropriate solution for multi-service VC. The data flows may consume a large amount of the bandwidth and destroy the QoS of flows with stringent delay constraint (e.g. voice). The Priority of voice packets over data packets is suitable for voice flows, but it is not appropriate for data flows especially when voice traffic is very important and may consume all the available bandwidth. A tradeoff between these two policies should be implemented. We should take into account a fair share of the bandwidth between different services and the time constraint of each service. WRR (Weighted Round Robin) and EDF (Earliest Deadline First) are two proposed algorithms to arbitrate between the data units that are ready for transmission on a link. The WRR algorithm consists of serving a predefined number of packets from each queue in a periodic mechanism. This number of packets divided by the cycle length of the algorithm represents the weight assigned to the queue. The basic idea of the EDF scheduling is to assign a deadline to every packet as it arrives, by adding the delay guarantee associated with the corresponding connection to the arrival time of the packet. The EDF scheduler selects the packet with the smallest deadline for transmission on the link. This scheme implies a dynamic priority as the priority of a packet increases over time spent in the system. Since all flows on the Iub and Iur interfaces become delay sensitive because of the air interface synchronisation, the AAL2 layer must guarantee the delivery delay of the traffic transported between the UTRAN nodes. Thus, once the traffic passes trough the air interface, the wired network should not be the bottleneck. The AAL2 connections must be well dimensioned in order to guarantee the delay requirements. The Timer-CU is an important parameter of the AAL2 protocol. Its value is critical in the case of low bit rate traffic. A very small Timer-CU value leads to a poor bandwidth utilization. A very high Timer-CU value optimises the bandwidth utilization, but leads to a higher packetization delay and consequently to a deterioration in the quality of service of the transported traffic. An optimal Timer-CU value should be chosen carefully to optimise the bandwidth utilization and keep an acceptable delay for the transported packets. The mini-cell transport as defined by the AAL2 protocol leads to the idea of AAL2 minicell switching. The main difference with the ATM switching is that the switched entities have variable lengths. Within the first UTRAN networks, the choice between an ATM switch and an AAL2 switch will be an important issue. For example, in the case of a D-RNC, all AAL2 connections in the same ATM VC are not directed to the same S-RNC. In this case, an ATM switch is not sufficient and the AAL2 switching technology is necessary in the D-RNC. However, in the case of a traffic concentrator between several Node Bs and a RNC, the switching entity may be an ANN. TÉLÉCOMMUN., 58, n 7-8, /25

9 1809-Her/Telecom 58/7-8 4/08/03 13:31 Page 9 R. MAKKÉ PERFORMANCE EVALUATION OF THE AAL2 PROTOCOL WITHIN THE UTRAN 9 ATM switch or an AAL2 switch. In this case, the choice between the two switching technologies will be based on a trade-off between performance, implementation complexity and cost of each solution. In this paper, we consider a concentrator node between several Node Bs and one RNC and we compare the performance of an AAL2 switch and an ATM switch. In order to evaluate the capacity of an ATM VC supporting an AAL2 traffic, we define the equivalent bandwidth of each flow as the mean bit-rate of the flow at the ATM level including all overheads. In other words, the equivalent bandwidth of an AAL2 connection is defined as the ratio of the total ATM bit-rate of all active AAL2 connections in the VC and the number of these connections. In order to guarantee the delay requirements of the AAL2 flows, the sum of all equivalent bandwidths of all active AAL2 connections must be less than or equal to a certain threshold that is a percentage of the PCR (Peak Cell Rate) value of the VC. This threshold depends on the type of flows and the RCR value. VI. TRAFFIC MODEL The traffic pattern at the AAL2 level is different from its pattern at source level. In fact, data flow goes through different protocol layers [11] and its characteristics change before entering the AAL2 layer. In order to have an accurate traffic model at the AAL2 level, we should analyse the behaviour of the traffic coming from upper layers. This traffic goes through a protocol stack which has the architecture described in Figure 6. The RLC is transparent for voice flows. For data flows, the RLC layer segments higher layer data units into a number of RLC-PDUs and sends them to the MAC layer. The MAC layer puts each RLC-PDU in one TB and selects the TBs that shall be sent in the same TTI on the air interface. The FP layer assembles all the TBs transmitted in the same TTI in one frame called FP- PDU. This FP-PDU is transmitted to the AAL2 layer. FIG. 6 Protocol stack architecture Architecture de la pile protocolaire. 9/25 ANN. TÉLÉCOMMUN., 58, n 7-8, 2003

10 1809-Her/Telecom 58/7-8 4/08/03 13:31 Page R. MAKKÉ PERFORMANCE EVALUATION OF THE AAL2 PROTOCOL WITHIN THE UTRAN In this paper, we consider two types of flows : voice (AMR: Adaptative Multi Rate) and data flows (UDD: Unconstrained Delay Data). At the source level, a voice model is defined by the 3GPP for the AMR codec [16,17]. This model consists of an ON/OFF model with exponential distribution of the ON and OFF periods. The time interval between two packets and the packet size are constant values. A SID (SIlence Descriptor) is sent immediately after the ON period to inform the receiver about the beginning of an OFF period. The SID is then sent during the OFF period only when the background noise level is changed in order to inform the receiver about this change. In our model, SID is sent every 160 ms during the OFF period. Table I represents the model parameters. TABLE I Voice model parameters. Paramètres du modèle de voix. Parameter Distribution Value ON period length Exponential 3 sec OFF period length Exponential 3 sec Time-interval between two packets Constant 20 ms AMR Packet size Constant Depending on the AMR coding mode SID packet size Constant 39 bits There are different AMR coding mode. Table II represents these different modes with their corresponding packet size. In this paper, the AMR 12.2 coding mode is used. The traffic coming from the voice source passes through the protocol stack described above. The traffic pattern is slightly changed when entering the AAL2 layer. In fact, the RLC protocol is transparent for voice traffic but at the MAC layer, one voice packet is transmitted on the radio channel in each TTI. The TTI value for voice connections is 20 ms. Thus, the FP layer receives one voice packet each 20 ms. It adds the FP overhead and sends this packet to TABLE II AMR coding modes. Les modes du codeur AMR. AMR coding mode Throughput Packet size AMR kbps 244 bits AMR kbps 204 bits AMR kbps 159 bits AMR kbps 148 bits AMR kbps 134 bits AMR kbps 118 bits AMR kbps 103 bits AMR kbps 95 bits ANN. TÉLÉCOMMUN., 58, n 7-8, /25

11 1809-Her/Telecom 58/7-8 4/08/03 13:31 Page 11 R. MAKKÉ PERFORMANCE EVALUATION OF THE AAL2 PROTOCOL WITHIN THE UTRAN 11 the AAL2 layer. The traffic entering the AAL2 layer is similar to the source traffic model with a different packet size (native packet size + MAC and FP headers). Ten signaling bytes are sent every 300 ms. For the data traffic, we consider in our study UDD flows which represent web browsing sources. A web browsing traffic model is defined by the 3GPP [17] : a web browsing session consists of a sequence of packet-calls corresponding to the download of pages. The number of packet-calls in a session is a geometrically distributed random variable with a mean of 5. Each packet-call represents the download of an internet file that has a Pareto with cut-off distributed size with a minimal file size of 1858 bytes and a maximal file size of bytes. The shape parameter of the normal Pareto distribution function is α = 1.1. the packet-calls are separated by a time interval called reading-time which is a geometrically distributed random variable with a mean of 12 seconds. This model is applicable on the Iu interface but it is not directly applicable on the Iub and Iur interfaces. In fact, RRC flow control mechanisms shape the traffic coming from upper layers so that the data throughput on the air interface does not exceed the defined bit-rate for the chosen UDD mode. Furthermore, the TTI parameter gives a periodic pattern to the traffic transported on radio channels and entering in the AAL2 layer. The RLC protocol splits upper layer packets into TBs with a predefined size and adds a 2-byte header. In each TTI, the MAC layer sends a certain number of TBs on the radio channel without exceeding the bit-rate of the UDD mode (in our model). This number is determined by a dynamic resource allocation algorithm at the RRC layer. The RRC is simplified in our model and we consider that the number of TBs to be sent within a TTI is fixed and the RLC-PDU size is also predefined. At the FP layer, all TBs transmitted in the same TTI for one user are assembled in one FP-PDU that is transmitted to the AAL2 layer after adding the FP overhead. The UDD traffic entering the AAL2 layer has the pattern represented in Figure 7. In each TTI, one FP-PDU is received at AAL2 layer. The FP-PDU size depends on the UDD mode and on the number of TBs sent in the TTI. The number of TBs sent in each TTI depends on the radio link utilization : if the radio link is low loaded, the RRC algorithm increases the number of TBs allowed to each user in the TTI because there is a free bandwidth. If the radio link is very loaded, the RRC algorithm decreases the allowed number of TBs for each user in order to share the bandwidth between all data users. In our model, we considered UDD sources at the AAL2 level with a bit-rate corresponding to the maximum bit-rate of the UDD type. In fact, we consider that the bandwidth on the radio FIG. 7 UDD traffic pattern at AAL-2 layer. Comportement du trafic UDD au niveau AAL2. 11/25 ANN. TÉLÉCOMMUN., 58, n 7-8, 2003

12 1809-Her/Telecom 58/7-8 4/08/03 13:31 Page R. MAKKÉ PERFORMANCE EVALUATION OF THE AAL2 PROTOCOL WITHIN THE UTRAN interface is available and the AAL2 transport layer should not be a bottleneck. In our simulations, the TTI value used for data channels is 40 ms and the TB size is 40 bytes. Thus, each 40 there is an FP-PDU to be sent to the AAL2 layer. This packet has a size corresponding to the UDD type. For example, for UDD 64kbps sources, there are 8 TBs sent in each TTI. Each TB has a size of 42 bytes (40 bytes + 2 bytes header). All these TBs all assembled in one FP-PDU overloaded by a 3-bytes header. VII. SIMULATION MODEL We consider an ATM link between a Node B and a RNC. In the Node B, an AAL2 multiplexer is implemented in order to aggregate several AAL2 connections into one ATM VC. Each AAL2 connection corresponds to one radio channel. The different layers are implemented in the simulation model as shown in Figure 8. FIG. 8 Simulation model. Modèle de simulation. Two scenarios are considered in the simulations : two different mono-service VCs, one for each type of traffic, and one multi-service VC for both types of traffic. In the scenario of traffic concentration, all Node Bs are connected to the concentrator (ATM or AAL2 switch) through ATM VCs and all flows coming from Node Bs are aggregated and sent to the RNC via one ATM VC. Figure 9 represents the scenario of traffic concentration. ANN. TÉLÉCOMMUN., 58, n 7-8, /25

13 1809-Her/Telecom 58/7-8 4/08/03 13:32 Page 13 R. MAKKÉ PERFORMANCE EVALUATION OF THE AAL2 PROTOCOL WITHIN THE UTRAN 13 FIG. 9 Traffic concentration. Concentration de trafic. VIII. SIMULATION RESULTS We studied by simulation the impact of the Timer-CU on traffic performance, different scheduling algorithms at the AAL2 level, the equivalent bandwidth, the capacity of a VC with different PCR values and a comparison between AAL2 and ATM switching technologies. VIII.1. Timer-cu The Timer-CU is an important parameter of the AAL2 protocol. Its value may affect the performance of the traffic transported on an AAL2 link. In a low loaded VC, if the Timer-cu value is very large, it may lead to a long packetization delay and consequently to a QoS degradation. Very low Timer-cu value may lead to a poor bandwidth utilization. Its value should be chosen carefully to obtain a trade-off between QoS requirement and bandwidth efficiency. In [18], we studied in depth the impact of this parameter. In this paper, we will present the most important results. Figure 10 represents the filling ratio in the case of a mono-service VC supporting voice traffic. The filling ratio is defined as the ratio between the number of bytes used for data in the ATM cell and the number of bytes available for AAL2 traffic in each ATM cell (47 bytes). Figure 11 represents the 95-percentile delay and the Standard Deviation of delay (StdDev) 13/25 ANN. TÉLÉCOMMUN., 58, n 7-8, 2003

14 1809-Her/Telecom 58/7-8 4/08/03 13:32 Page R. MAKKÉ PERFORMANCE EVALUATION OF THE AAL2 PROTOCOL WITHIN THE UTRAN for voice packets. Figures 12 and 13 represent respectively the filling ratio and the 95-percentile delay in the case of a mono-service VC supporting data traffic. FIG. 10 Filling ratio for voice VC. Taux de remplissage pour un circuit virtuel de voix. FIG. 11 Delay and StdDev for voice packets. Délai et Ecart-Type pour les paquets de voix. ANN. TÉLÉCOMMUN., 58, n 7-8, /25

15 1809-Her/Telecom 58/7-8 4/08/03 13:32 Page 15 R. MAKKÉ PERFORMANCE EVALUATION OF THE AAL2 PROTOCOL WITHIN THE UTRAN 15 FIG. 12 Filling ratio for data VC Taux de remplissage pour un VC de données. FIG. 13 Delay for data packets Délai pour les paquets de données. We observe that for high Timer-CU values, the packetization delay decreases when the number of streams increases. The filling ratio increases with the number of streams in the case of a small Timer-CU value. At high load, the Timer-CU does not have an important impact on performance. In [18], we showed that the optimal Timer-CU value chosen in the case of mono-service VC is suitable for multi-service VC. The transfer delay depends on the value of the Timer-CU regardless the PCR value of the VC. On the other hand, the filling ratio depends on the Timer-CU value that is a function of the PCR value. However, the delay is the more critical parameter in UTRAN. Thus, a Timer-CU value chosen to satisfy delay requirements is needed. 15/25 ANN. TÉLÉCOMMUN., 58, n 7-8, 2003

16 1809-Her/Telecom 58/7-8 4/08/03 13:32 Page R. MAKKÉ PERFORMANCE EVALUATION OF THE AAL2 PROTOCOL WITHIN THE UTRAN VIII.2. Scheduling mechanisms In the case of a mono-service VC, all AAL2 connections have the same priority. Thus, the FCFS policy is suitable. Another possible algorithm is the RR (Round Robin). This algorithm is a cyclic mechanism that serves periodically one packet from each connection. A mono-service VC (CBR VC Constant Bit Rate, with PCR = 2 Mbps) supporting voice channels is considered and the two algorithms (FCFS and RR) are compared. The utilization of the VC depends on the number of voice streams (between 0.1 and 0.85). Figure 14 represents the 95-percentile delay in the case of the FCFS and RR algorithms and Figure 15 represents the standard deviation of delay (StdDev). FIG. 14 Voice delay (FCFS vs RR). Délai pour la voix (FIFO vs RR). FIG. 15 StdDev of delay (FCFS vs RR). Écart-Type du délai (FIFO vs RR). ANN. TÉLÉCOMMUN., 58, n 7-8, /25

17 1809-Her/Telecom 58/7-8 4/08/03 13:32 Page 17 R. MAKKÉ PERFORMANCE EVALUATION OF THE AAL2 PROTOCOL WITHIN THE UTRAN 17 We observe that the simple FCFS mechanism is very suitable in the case of a mono-service VC for voice streams. It gives the same performance of the RR mechanism. The same experience was repeated with different Timer-CU values and we observed the same result. In the case of a multi-service VC where voice channels and data channels are aggregated, the implementation of a scheduling algorithm in the CPS multiplexer is mandatory. In fact, data packets are more tolerant to delay than voice packets. If a voice packet and a data packet are presented at the CPS multiplexer, we can serve the voice packet first, then the data packet. The FCFS (or FIFO) policy is not an appropriate solution to arbitrate between voice and data packets because data flow may disturb the voice flow and consequently leads to a QoS degradation for real-time traffic. A scheduling mechanism based on the priority of the real-time flow over the non real-time flow gives better performance for voice traffic but increases considerably data delays. A trade-off between these two solutions is needed. In this paper, we studied the EDF algorithm (Earliest Deadline First) based on a transmission deadline assigned for each packet and the WRR algorithm (Weighted Round Robin) based on a weight assigned for each flow. These two alternatives may be a compromise if the deadlines and the weights are appropriately chosen. In our simulations, we considered different combinations of flows : 20% voice and 80% data 50% voice and 50% data 80% voice and 20% data Three Types of UDD traffic are considered : UDD 64 kbps, UDD 144 kbps and UDD 384 kbps. Different weights for the WRR algorithm are considered: 1/5 for voice and 4/5 for data 1/2 for voice and 1/2 for data 4/5 for voice and 1/5 for data Different deadlines for the EDF algorithm are considered : 2ms for voice and 20ms for data 2ms for voice and 50ms for data 5ms for voice and 100ms for data For each type of flow, the buffer size is supposed to be infinite which means that the system is without loss. A CBR VC with a PCR value of 2 Mbps is considered. The load of the VC is about 0.7. We calculated the complementary distribution of delay for voice and data packets. In this paper, we present few simulation results for some flow combinations and some UDD types. Figures 16, 17 and 18 represent the complementary distribution of delay for different flow combinations and for the UDD 64kbps mode. Figures 19,20 and 21 represent the same results for the UDD 384kbps mode. In special cases, WRR gives better performance than EDF. This result is very clear when the data flow has a very bursty pattern. For example, when the UDD 384kbps traffic is used, the FP-PDU has a greater size and consequently the number of mini-cells arriving at the same time in the CPS multiplexer is higher and the bursty pattern is clear. Thus, when a long data packet arrives in the AAL2 layer, it will be segmented into several minicells that have the same deadlines. When the CPS multiplexer gives access to the data queue, it will serve all data minicells before giving access to the voice queue. At this time, if a voice minicell arrives in the CPS multiplexer, it will wait for the end of service of all data mini-cells for which the deadline has expired, and the waiting time may be very large depending on the data source type and the 17/25 ANN. TÉLÉCOMMUN., 58, n 7-8, 2003

18 1809-Her/Telecom 58/7-8 4/08/03 13:32 Page R. MAKKÉ PERFORMANCE EVALUATION OF THE AAL2 PROTOCOL WITHIN THE UTRAN FIG. 16 Complementary distribution : 20% voice and 80% data UDD 64kbps Distribution complémentaire de probabilité : 20% voix et 80% données UDD 64kbps. FIG. 17 Complementary distribution : 50 % voice and 50% data UDD 64kbps Distribution complémentaire de probabilité : 50% voix et 50% données UDD 64kbps. data packet length. The WRR algorithm is a good solution in this case because it shares the bandwidth between all flows. The weights of WRR depend on the traffic configuration. It is not easy to determine these weights if we do not have an idea of the traffic entering the AAL2 layer. In the case where we know the traffic configuration (the percentage of the entering traffic : Pi) and the mean packet size of each service (Si), we may use these parameters with the maximum delay allowed for each service (Ti) in order to determine the weight of each queue (Wi) : ANN. TÉLÉCOMMUN., 58, n 7-8, /25

19 1809-Her/Telecom 58/7-8 4/08/03 13:32 Page 19 R. MAKKÉ PERFORMANCE EVALUATION OF THE AAL2 PROTOCOL WITHIN THE UTRAN 19 FIG. 18 Complementary distribution : 80% voice and 20% data UDD 64kbps Distribution complémentaire de probabilité : 80% voix et 20% données UDD 64kbps. FIG. 19 Complementary distribution : 20% voice and 80% data UDD 384kbps Distribution complémentaire de probabilité : 20% voix et 80% données UDD384kbps. FIG. 20 Complementary distribution : 50% voice and 50% data UDD 384kbps. Distribution complémentaire de probabilité : 50% voix et 50% données UDD 384kbps. 19/25 ANN. TÉLÉCOMMUN., 58, n 7-8, 2003

20 1809-Her/Telecom 58/7-8 4/08/03 13:32 Page R. MAKKÉ PERFORMANCE EVALUATION OF THE AAL2 PROTOCOL WITHIN THE UTRAN FIG. 21 Complementary distribution : 80% voice and 20% data UDD 384kbps. Distribution complémentaire de probabilité : 80% voix et 20% données UDD 384kbps. Wi Pi Tj Sj = Wj Pj Ti Si VIII.3. Equivalent bandwidth and vc capacity The capacity of a VC is an important issue. In fact, the simultaneous number of the AAL2 connections in an ATM VC with a given PCR value should be controlled to guarantee the required quality of service for the traffic supported by the AAL2 connections. The equivalent bandwidth as defined above gives an idea of the capacity needed for each AAL2 connection. In order to guarantee the delay requirements, the sum of the equivalent bandwidths for all active connections must be less than or equal to a certain threshold. This threshold is a percentage of the PCR value of the VC. This percentage factor is computed by simulation for different PCR values and different flow types. The equivalent bandwidth may be computed analytically or by simulation. Equations 1 and 2 give the equivalent bandwidth for AMR flow and UDD flow respectively. Equation 1 : BW = D 8 + O_MAC + O _FP + O _CPS T 8 53 TI 47 ON ON + OFF Equation 2 : BW = FP_PDU_size 45 O_CPS + FP _PDU_size T 8 53 TI 47 ON ON + OFF ANN. TÉLÉCOMMUN., 58, n 7-8, /25

21 1809-Her/Telecom 58/7-8 4/08/03 13:32 Page 21 R. MAKKÉ PERFORMANCE EVALUATION OF THE AAL2 PROTOCOL WITHIN THE UTRAN 21 where : FP_PDU_size = D 8 RLC_PDU_size BW : equivalent bandwidth. (RLC_ PDU_size+ O_ RLC+ _ MAC) + O_ FP D : source bit-rate [bps]. For AMR traffic, D is the bit-rate of the AMR mode used. For UDD traffic, D is bit-rate of the corresponding mode. TTI : Transmission Time Interval of the channel [sec] O_MAC, O_FP, O_CPS : Overhead size for MAC, FP and CPS layer respectively. ON, OFF : mean value of ON and OFF periods [sec]. 53 : size of an ATM cell with header. 47 : number of bytes in an ATM cell offered to the AAL2 traffic. RLC_PDU_size : the size of the RLC-PDU predefined by the RRC functions. x : is the first integer greater than or equal to x. The equivalent bandwidth computed by simulation is slightly greater than the value given by these equations. In fact, the ATM cells may be filled by padding if the Timer-CU expires before the traffic fills the ATM cells. The output throughput is greater than the calculated equivalent bandwidth because of the additional bandwidth added by the padding bytes. Figures 22, 23, 24 and 25 represent the equivalent bandwidth and the maximum utilization of the VC (the threshold) for voice traffic (AMR12.2) and UDD traffic (UDD64) respectively. FIG. 22 Equivalent bandwidth for AMR12.2 traffic. Bande passante équivalente pour le trafic AMR /25 ANN. TÉLÉCOMMUN., 58, n 7-8, 2003

22 1809-Her/Telecom 58/7-8 4/08/03 13:32 Page R. MAKKÉ PERFORMANCE EVALUATION OF THE AAL2 PROTOCOL WITHIN THE UTRAN FIG. 23 Threshold for AMR12.2 Seuil pour AMR12.2. FIG. 24 Equivalent bandwidth for UDD64 traffic. Bande passante équivalente pour le trafic UDD 64. FIG. 25 Threshold for UDD64. Seuil pour UDD64. ANN. TÉLÉCOMMUN., 58, n 7-8, /25

23 1809-Her/Telecom 58/7-8 4/08/03 13:32 Page 23 R. MAKKÉ PERFORMANCE EVALUATION OF THE AAL2 PROTOCOL WITHIN THE UTRAN 23 The equivalent bandwidth is stable for the VCs with a PCR greater than 1 Mbps. It is about 10 kbps for AMR12.2 traffic and about 9 kbps for UDD64 traffic. It is clear that for low PCR values, the maximum utilization of the VC is small especially for UDD traffic. In fact, UDD traffic is very bursty and at low PCR values, we should keep a large bandwidth free in order to be used by the instantaneous bursts. In this case, the maximum VC bandwidth is used only in a short time period when several bursts are active at the same time. The rest of the time, the VC capacity has a poor utilization. When the PCR value increases, the free bandwidth needed becomes lower and thus, the threshold increases and reaches 82% for AMR12.2 traffic and 63% for UDD64 traffic. VIII.4. Comparison between AAL2 and ATM switching technologies The performance of the AAL2 switching technology is treated in this section in the case of a concentration point between several Node Bs and one RNC. The concentration point is FIG. 26 TAAL2 and ATM switching. Commutation AAL2 et ATM. 23/25 ANN. TÉLÉCOMMUN., 58, n 7-8, 2003

24 1809-Her/Telecom 58/7-8 4/08/03 13:32 Page R. MAKKÉ PERFORMANCE EVALUATION OF THE AAL2 PROTOCOL WITHIN THE UTRAN considered to be an AAL2 switch or an ATM switch in order to compare these two switching alternatives. All the CBR VCs coming from the Node Bs to the switch are aggregated in the same output CBR VC. The PCR value of each VC entering the switch is 2 Mbps. The PCR value of the VC between the switch and the RNC is 2 Mbps. The load of a VC between the Node B and the switch is 13% (low loaded VC). We consider two types of traffic : AMR12.2kbps and UDD64kbps with a combination of 50% voice and 50 % data. The scheduling mechanism used in the AAL2 multiplexer is the absolute priority of voice packets over data packets. We consider that the links between the Node Bs and the switch are low loaded in order to study the impact of the AAL2 switching on the multiplexing gain. Figure 26 represents the percentile delay for voice and data packets in the case of an AAL2 switch and an ATM switch. This is the delay between the SAP (Service Access Point) of the AAL2 layer in the Node B and the other SAP of the AAL2 layer in the RNC including switching delay. We observe that in the case of an AAL2 switch, we can aggregate a larger number of Node Bs than the case of the ATM switch with acceptable delay. In fact, when multiplexing at the AAL2 layer, we can differentiate between different types of flows and implement a scheduling mechanism in order to give priority for stringent delay packets (e.g. voice packets). Furthermore, in the case of an AAL2 switch, we may benefit from the CPS multiplexer to eliminate the padding in the cells coming from the low loaded VCs and reduce the bit-rate of the outgoing flow. In the case of an ATM switch, the bit-rate of the outgoing flow is the sum of all incoming bit-rates. If the entering VCs are high loaded, the AAL2 switching technology will not have an important advantage and an ATM switch is sufficient. IX. CONCLUSION In this paper, we presented some issues in performance of the AAL2 protocol studied by simulation. The Timer-CU is a very important parameter of the AAL2 protocol. A high Timer- CU value leads to high packetization delay and consequently to a QoS degradation especially in the case of low loaded VCs. A small Timer-CU value leads to a poor bandwidth utilization. A trade-off between delay and bandwidth efficiency may give the optimal value of the Timer- CU and this value may be chosen in the case of mono-service VC. In the case of low loaded VC between a Node B and a RNC, the Timer-CU may not have an important impact because we can choose a small value in order to guarantee the packetization delay and we do not care about the bandwidth efficiency because the VC is low loaded. But this is not true if there is a concentration node between the Node B and the RNC, especially if this node is an ATM switch. In fact, if the VCs entering the ATM switch are low loaded and if we choose a small Timer-CU value in order to guarantee the packetization delay, the ATM cells will be partially filled. This is not a problem for a VC entering the concentrator because it is low loaded but the output VC may be high loaded and since in the ATM switch there is no multiplexing at the AAL2 level, the cells in the output VC will be partially filled and consequently lead to a bandwidth loss. In this case, we should carefully choose the value of this timer. A Timer-CU between 1 ms and 2 ms may be an optimal value. The scheduling mechanism is very important in the CPS multiplexer especially in the case of multi-service VC. We showed that the FCFS (or FIFO) mechanism is appropriate for mono- ANN. TÉLÉCOMMUN., 58, n 7-8, /25

25 1809-Her/Telecom 58/7-8 4/08/03 13:32 Page 25 R. MAKKÉ PERFORMANCE EVALUATION OF THE AAL2 PROTOCOL WITHIN THE UTRAN 25 service VCs and gives good performance. Furthermore, it is less complicated than the Round Robin algorithm which requires a timer management for the periodic mechanism. In the case of multi-service VCs, FCFS is not an appropriate solution. Priority penalizes data traffic especially when voice traffic is heavy. EDF and WRR are a trade-off between these two solutions. The implementation of the EDF algorithm is more complicated than WRR because it requires the management of several timers, while WRR requires only the management of one periodic timer for all queues. Furthermore, EDF may lead to a QoS degradation for voice traffic in some specific situations as showed above. WRR may be used as a scheduling mechanism in the CPS multiplexer and the weights should be chosen appropriately. The equivalent bandwidth is an important parameter which gives an idea of the bandwidth needed by each AAL2 connection and of the maximum number of AAL2 connections that can be supported by a given VC with delay guarantee. Finally, the problem of the AAL2 switching is considered and we concluded that when switching at the AAL2 layer, we can obtain a large multiplexing gain especially when we have concentration points. Within the framework of release 5 of the 3GPP, the IP solution has been selected as another transport technology within UTRAN [17]. Many manufacturers and operators defend this solution in the perspective of the deployment of global IP networks. This technology becomes very attractive if future data traffic represents a significant part of the whole traffic transported by the network. The IP transport technology in UTRAN is the subject of our current studies. Manuscrit reçu le 3 octobre 2002 Accepté le 10 février 2003 REFERENCES [1] ITU-T I.363.2, B-ISDN ATM Adaptation Layer Specification: AAL Type 2. [2] ITU-T I.366.1, Segmentation and Reassembly Service Specific Convergence Sublayer for the AAL type 2. [3] 3GPP website, [4] MINICEL PROJECT, ENST-Paris, France Télécom R&D, Mitsubishi Electric ITE, Technical Reports 1,2,3,4,5,6,7,8,9. [5] ITU-T I.366.2, AAL type 2 service specific convergence sublayer for narrow-band services. [6] ETSI TS , UTRAN Overall Description. [7] ETSI TS , UTRAN Iu Interface RANAP Signalling. [8] 3G TS , UTRAN Iub Interface User Plane Protocols for Common transport Channel Data Streams. [9] 3G TS , UTRAN Iub/Iur Interface User Plane Protocol for DCH Data Streams. [10] ETSI TS , UTRAN Iur Interface General Aspects and Principles. [11] 3G TS25.301, Radio Interface Protocol Architecture. [12] 3G TS , RLC Protocol Specification. [13] 3G TS25.331, Radio Resource Control (RRC) Protocol Specification. [14] 3G TS25.321, MAC Protocol Specification. [15] 3G TS23.107, QoS Concept and Architecture. [16] ETSI TR , Selection Procedure for the choice of radio transmission technologies of the UMTS. [17] 3G TR , IP Transport in UTRAN. [18] MAKKÉ (R.), TOHMÉ, (S.), COCHENNEC (J-Y.), PAUTONNIER (S.), Impact of the Timer-CU of the AAL2 Protocol on Traffic Performance within the UTRAN, IFIP Personal Wireless Communications 2001, 8-10 August 2001, Lappeenranta, Finland, pp /25 ANN. TÉLÉCOMMUN., 58, n 7-8, 2003