WiMAX: MAC Layer Performance Assessments

Transcription

1 WiMAX: MAC Layer Performance Assessments A. Bestetti,G.Giambene, S. Hadzic Alcatel-Lucent, Via Trento, 30, I Vimercate, Milano, Italy University of Siena, Via Roma, 56, I Siena, Italy Abstract This paper focuses on WiMAX since this wireless technology allows broadband communications. In particular, we provide here a preliminary study on MAC layer performance as well as a sensitivity study to its parameters. A simulation approach has been adopted, based on the ns-2 environment to investigate the impact that real-time traffic has on non-real-time traffic and vice versa. The study presented in this paper has been carried out in the framework of the EU FP6 WEIRD project. I. INTRODUCTION Broadband wireless access systems are gaining now momentum everywhere for their capabilities to allow users to have a high-speed connection to the Internet. The IEEE d standard allows broadband wireless access for fixed users, while the IEEE e amendment supports fixed and mobile users. IEEE d defines different air interface options; we focus here on the Orthogonal Frequency Division Multiplexing (OFDM) case on the basis of one scenario defined in the WEIRD (WiMAX extension to Isolated Research Data Networks) European FP6 project [1]. OFDM is a digital modulation technique that combats frequency-selective fading by splitting the transmission flow on parallel orthogonal flat narrowband channels, named sub-carriers [2],[3]. IEEE d uses an FFT with 256 orthogonal sub-carriers. Each transmitted OFDM symbol has a Cyclic Prefix (CP) that completely eliminates Inter-Symbol Interference (ISI) as long as the CP duration is longer than the channel delay spread. An OFDM symbol is made up of three different types of sub-carriers: data, pilot and null. OFDM allows sub-carriers to be adaptively modulated depending on distance and noise level. The following modulation and coding combinations are available: BPSK 1/2, QPSK 1/2, QPSK 3/4, 16QAM 1/2, 16QAM 3/4, 64QAM 2/3, and 64QAM 3/4. With Worldwide Interoperability for Wireless Microwave Access (WiMAX) we mean different possible standard versions within the IEEE family. WiMAX defines the Medium Access Control (MAC) framework, traffic classes and signaling, but it does not identify a specific scheduling algorithm to manage the different traffic classes [4] - [6]. This paper deals with an evaluation of the MAC layer performance considering the impact of the access phase and the scheduling for a configuration with non-real-time and real-time traffic flows. Correspondence to: G. Giambene ( giambene@unisi.it). This work has been carried out with the support of Alcatel-Lucent in the framework of the EU FP6 WEIRD project (Integrated Project n ). II. WEIRD PROJECT OVERVIEW WEIRD is an IST FP6 project with the scope to leverage the features offered by broadband wireless connectivity and to test and validate actual wireless state-of-the-art technology, thus interacting in a convergent heterogeneous network architecture in order to prepare for the deployment of next-generation information and communications networks across Europe. The main objective of the WEIRD project is to have a seamless E2E integration of various network technologies, to provide Quality of Service (QoS) over wireless communication channels, and to achieve a wireless connection for the last mile by using the IEEE standard technology (WiMAX), overcoming the cost barriers of wired technologies. One of the main WEIRD tasks relies upon implementing four test beds located in different European countries to test WEIRD solutions over large-scale experimentations in real settings. A scenario envisages the wireless real-time data links between seismological sensors deployed over a volcano and the central site of the Osservatorio Vesuviano in Italy or of the Icelandic Meteorological Office. These real-time data can be used to analyze volcano status and evolution and to share information with the scientist community all over the Europe. WEIRD tasks include deploying a proprietary network architecture supporting different applications (with related QoS levels) to collect information coming from distributed terminals located in critical and isolated zones. These services will be shared with the scientist community across the Europe with connectivity granted by the GEANT pan-european network and relevant national research networks. III. WIMAX MAC LAYER The IEEE standard defines a connection-oriented MAC and the signaling mechanism for information exchange between Base Station (BS) and Subscriber Station (SS). The MAC layer takes packets (called MAC Service Data Units, MSDUs) from the upper layer and organizes them into MAC Protocol Data Units (MPDUs) for transmission. Connection ID (CID) is the identifier of a unidirectional connection between MAC peers over the air interface. A Service Flow (with related Service Flow ID, SFID) is a MAC transport layer channel for unidirectional (uplink or downlink) transport of MPDUs within a QoS class for a connection. WiMAX uses a variable-length MPDU (max length of 2048 bytes) formed by a 6-byte Generic MAC Header (GMH), a payload of variable length and an optional CRC. Multiple

2 MPDUs may be concatenated into a single burst to save physical overhead. Similarly, multiple MSDUs from the same higher-layer service may be packed into a single MPDU to save MAC header overhead. Large MSDUs may be fragmented into smaller MPDUs and sent with multiple bursts. In the case of Time Division Duplexing (TDD) air interface, time is divided into downlink and uplink sub-frames with movable boundaries and a gap interval in-between. Each frame has a fixed duration of a few milliseconds. The downlink subframe comes first and then there is the uplink part. The first part of the downlink phase contains a preamble and a lot of broadcast signaling information to coordinate the usage of resources (e.g., DL-MAP and UL-MAP messages contain the burst allocation decided by the BS, respectively for downlink and uplink). Two initial parts of the uplink sub-frame are used for contention-based transmissions for initial ranging of newly joining SSs and for best effort bandwidth requests of alreadyassociated SSs. Note that 1 symbol is the preamble overhead of each uplink burst (short preamble). IV. WIMAX RESOURCE MANAGEMENT AND QOS SUPPORT Four MAC components are necessary to manage QoS, such as: admission control, scheduling, buffer management, and access scheme. In order to support different applications, WiMAX provides the following QoS traffic classes: UGS (Unsolicited Grant Service), ertps (Extended Real Time Polling Services - from e), rtps (Real Time Polling Service), nrtps (Non Real Time Polling Service), BE (Best Effort). The BS decides about the transmission in uplink and downlink. In downlink, only the BS transmits, but in uplink all SSs can transmit. Uplink resource allocation for packet transmission is performed by the BS through signaling to the SSs. At the SS, the packet scheduler will retrieve the packets from the queues and transmit them to the BS in the appropriate assigned resources, defined by the UL-MAP sent by the BS. The role of the scheduler (both BS side and SS side) is to transfer packets from the queues of the connections to the bursts, to determine the quantity of packets that will be sent in the bursts and also the fragmentation needed. All packets from the application layer are classified on the basis of CID and SIF and are forwarded to the appropriate queue (i.e., UGS, rtps, ertps, nrtps, and BE). The resource allocation to SSs (uplink direction) is based on a request-grant mechanism [7]. Bandwidth requests are always per connection. The standard defines the resource allocation methods (i.e., grant mechanisms) below for SSs transmissions to BS; these mechanisms can be mapped to the different traffic classes. Unsolicited bandwidth grants. The SS sends a request only once, and the BS will start sending grants to allocate resources periodically. This is suitable for UGS class. Unicast polling. This allocation technique is used when bandwidth resource demand is not relevant enough to use unsolicited bandwidth grants for SSs. When an SS is polled, resources (communicated via UL-MAP message) are allocated in uplink to allow the SS to send a bandwidth request message. If the SS does not need to make a request (decision of the SS scheduler), it may use this bandwidth allocation to send data. If SS does not have data to transmit it can use padding CID to indicate that it is not sending the bandwidth request. This access method is suitable for rtps, ertps, and nrtps classes. Contention-based polling. If there is not sufficient bandwidth to poll individually SSs and if some SSs can be inactive, a certain amount of every uplink sub-frame (called contention part) is allocated for sending contention-based requests. This access mechanism is suitable for the BE traffic class. Contention resources can be used by either all SSs (broadcast polling) or a group of SSs (multicast polling). The access protocol used for the contention phase is described below. Bandwidth requests can be standalone request (for unicast or contention-based polling) or use the piggybacking mechanism. In each frame, there is in the uplink part a contention phase that is organized according to transmission opportunities (the transmission of one bandwidth request needs a transmission opportunity). Because the uplink burst profile can change dynamically, all requests for bandwidth shall be made in terms of the number of bytes needed to carry the GMH and payload. As soon as an SS wants to send a bandwidth request through contention (i.e., when this SS has a new packet in its buffer and there is no other pending request), it enters the contention resolution algorithm. The initial backoff and maximum backoff window values are selected by the BS and are specified in the UCD message. They are given as a power of two; for example, the value of 4 indicates a backoff window value At the first attempt, the backoff value is equal to the initial one. The access algorithm operates as follows: the SS will select a random number in the interval (0, backoff window) that will be used to indicate the number of transmission opportunities that will be missed in the contention phase. Then, the SS transmits the bandwidth request on a suitable transmission opportunity and waits for a given number of subsequent UL-MAP messages (timer T16) to receive a data transmission grant from the BS (if the request has been received without collisions and has been satisfied). If timer T16 expires without an allocation, the SS doubles the backoff window (as long as it is less than the maximum value) and repeats the contention procedure. This retry process continues until the maximum number of retries (Request Retries limit = 16) has been reached. If this happens, the SS discards the bandwidth request. The contention phase depends on the following parameters: the contention size c in transmission opportunities, that is the number of transmission opportunities per frame; the minimum and the maximum value of the backoff window, W min and W max ; timer T16; the Request Retries limit; the frame duration, T frame. The bigger the c value, the lower the risk of collisions, but the lower the resources for data traffic, thus increasing the transmission delay. Note that the

3 optimal number of c maximizing the throughput of successful access attempts per transmission opportunity should be equal to the number of contending requests in each frame. A too high T16 value could cause too delay to react to collisions. On the other hand, if T16 is too small (the minimum T16 value is 10 ms), unuseful reattempts could be triggered for requests successfully received at the BS, but not yet served due to the service delay (congestion of resources). With the following condition, we have that T16 expires after the first attempt has been accomplished (. is the ceiling function): ( 1+ Wmin ) T frame. (1) T 16 c V. SIMULATOR CHARACTERISTICS In this study, we have considered the WiMAX model (TDD version) for the ns-2 environment developed by NIST (National Institute of Standards and Technology, US) [8]. A. PHY Layer The NIST model physical layer is based on OFDM (FFT with 256 values). There are 192 data sub-carriers (an SS in uplink can use all of them), and the symbol duration and the slot duration can be determined as [3]: OFDM symbol duration = T g + T u = N FFT s f BW (1 + G), (2) 4 physical slot duration = s f BW, (3) where T u denotes the useful symbol duration (the reciprocal of the sub-carrier bandwidth), T g is the CP duration (typically, the CP phase is expressed in terms of the ratio G = T g /T u ), and s f is the sampling factor. Moreover, we consider the parameter values in Table I that refer to the 3.5T1 WiMAX profile. Hence, on the basis of (2) with N FFT = 256 (the number of sub-carriers), we obtain OFDM symbol duration = [s] and physical slot duration = [s]. From (2) and (3), one symbol requires N FFT (1 + G)/4 slots. We can derive the number of slots and hence the number of symbols available in a frame for data transmissions considering the presence of RTG and TTG gaps for switching between uplink and downlink parts and vice versa (each of 20 slots); we have: Number of data symbols/frame = = 4 N FFT(1+G) [ ] T frame physical slot duration RT G TTG. (4) On the basis of (4), we have 124 OFDMA symbols per frame, being the symbol the minimum granularity in the resource allocation process. These 124 symbols are used for all the types of downlink and uplink traffic flows (i.e., broadcast downlink management messages, unicast management messages, data messages). In our study, we have considered a configuration with downlink/uplink resource ratio as 0.3/0.7 in order to allow a higher uplink capacity for the transmission of uplink time-critical data. Hence, there are 37 symbols for downlink and 87 symbols for uplink. The uplink and the downlink resources for data traffic are further reduced due to the resources needed for management messages. Hence, for downlink we have to consider DL PREAMBLE, DL MAP and UL MAP and gaps between the different parts, totally needing 10 symbols in the simulator. Moreover, for uplink, 19 symbols are used for initial ranging procedures of SSs and a variable number of symbols are used to accommodate the variable number of c transmission opportunities per frame (the BPSK 1/2 mode is used); we are referring here to the region request full contention method and sub-channelization is not used. At the beginning of the ranging phase and the contention one there is one gap symbol. According to NIST simulator settings [8], we have that the number of OFDM symbols used for the contention phase is S c = S c = c + x, where x is the smallest integer number so that x 96 t 48 c and t = 8 bits denotes a trailer byte. Hence, c = 2 requires S c =4 OFDM symbols and c = 5 needs S c = 8 OFDM symbols 1. With our settings, practically, we have N data sym, downlink = 27 symbols that can be used for downlink data traffic and N data sym, uplink = S c 2 = 66 S c symbols that can be used for uplink data traffic. Note that these resources for data traffic are also used for management connections (basic, primary and secondary connections corresponding to each active SS) that constitute a negligible traffic load. The uplink Cap up and the downlink Cap down capacities in bit/s for information traffic can be evaluated (upper bound) as follows: Cap = N data sym 192 log 2 (M) r, (5) T frame where M is the number of data sub-carriers and r denotes the RS-CC code rate. Moreover, in the uplink case we consider Cap = Cap up and N data sym = N data sym, uplink and for the downlink we use Cap = Cap down and N data sym = N data sym, downlink. Correspondingly, for instance for BPSK 1/2, we have kbit/s in downlink and 1113 kbit/s in uplink with c =5.Note that these capacities are further reduced due to the presence of preamble bursts. Finally, it is important to note that modulation and coding adaptation is not supported in the simulator; a threshold mechanism is used to determine the cell size. B. MAC Layer Outgoing packets are categorized by the classifier object only on the basis of their destination and are then sent to the connection object. When two MAC nodes (SS and BS) establish a new connection, the simulator creates two objects of type Connection, one outgoing (at SS side) and one incoming (at BS side), or vice versa. The simulator defines GMH with all the fields as specified in the standard. The NIST simulator currently supports only contention-based polling (broadcast, not multicast) for sending bandwidth requests. Note that 1 In the simulator, a transmission opportunity is the resource needed to send a bandwidth request burst formed by a short preamble (1 OFDM symbol) plus a GMH plus a trailer. This is slightly different with respect to the standard, where a bandwidth request is totally formed by 2 OFDM symbols.

4 piggybacking, ARQ and packing are not implemented. The BS scheduler determines the composition of sub-frames that are practically organized as in the standard. Several parameters can be set from Tcl scripts, such as frame length, the ratio between downlink and uplink sub-frame durations and the contention size. Contention zones are always at the beginning of uplink sub-frame; each request is preceded by the preamble (1 OFDM symbol for bandwidth requests, and 2 OFDM symbols for ranging) [2]. The contention resolution algorithm is implemented as specified in the standard. The NIST module implements both BS and SS schedulers; however, there is no traffic differentiation: only BE traffic class is supported and the scheduling of the different connections is based on a round robin approach, as described below. In downlink, the BS scheduler looks into each peer node and tries to schedule data from its connections to the downlink sub-frame. Connections in one peer node are checked in the following order: i) Basic connection; ii) Primary connection; iii) Secondary connection; iv) Data connection. The service of connections is performed in a round robin way. As long as there is free space in the downlink sub-frame, the scheduler will assign a data burst to the connection. Once the free space in the downlink sub-frame is over, the scheduler will remember where it has stopped, and in the next frame it will continue to assign bursts from there. In uplink, the BS scheduler will first reserve space for contention slots. As said before, information about the bandwidth needs is stored in the peer node objects in the MAC layer. When scheduling uplink transmissions, the scheduler will look at this information from connections, and will use a round robin procedure as for the downlink part. During this process, it will also add 1 OFDM symbol (= short preamble) between bursts. The SS scheduler is responsible for taking the data transmission opportunities allocated by the BS and for assigning them to the appropriate incoming connections. If bandwidth is assigned to a given connection, the SS scheduler will also fill the bursts with the data from its outgoing connections. If the SS has not an outstanding request, the SS scheduler looks at all outgoing connections and generates bandwidth requests, sent in the contention phase. Each request is of the aggregated type. VI. PERFORMANCE EVALUATION In this study, we consider a scenario with inelastic and elastic traffic flows respectively for video and data transfer applications. We refer to the configuration in Fig. 1 with a variable number N of SSs receiving FTP downlink flows and N SSs transmitting video-cbr uplink flows to the BS. Thus, we have three traffic flow types: FTP downlink data packets, uplink video-cbr packets and uplink TCP ACKs. The video stream is with Standard Interchange Format (SIF, pixels) resolution: one IP packet of 470 bytes is generated every 13 ms (corresponding to about 282 kbit/s). The N video-cbr flows and the N FTP flows are started at Fig. 1. WiMAX network scenario considered. TABLE I SIMULATION PARAMETER VALUES (3.5T1 WIMAX PROFILE). Parameter value BS DL/UL ratio of the TDD frame 0.3/0.7 Frame duration, T frame 5ms Channel bandwidth, BW 7MHz Receive/Transmit gap 20 physical slots Cyclic prefix, G = T g/t u 0.25 Queue length of each data connection 100 packets DCD/UCD interval 5s Channel number n. 1 at GHz Modulation and coding BPSK 1/2 or 16QAM 1/2 Sampling factor for 7 MHz BW, s f 8/7 Transmission power (both BS and SS) 27 dbm Maximum backoff window, W max 63 trans. opport. randomized instants (uniformly distributed in the interval 0-30 ms) to avoid phase effects with the WiMAX frame. For the FTP traffic, we use an ACK-cloked TCP traffic model based on TCP NewReno with packets of 1500 bytes generated at the IP level. Simulator settings are shown Table I and refer to the 3.5T1 WiMAX profile [9]; in addition to this, all wired links in Fig. 1 are at 100 Mbit/s with 1 ms of propagation delay and droptail buffer policy. In this preliminary study, all the traffic flows belong to the BE class. We are interested to evaluate the WiMAX performance expressed in terms of mean delay and mean goodput for video CBR flows and mean TCP goodput for FTP flows. We aim to show the impact on performance due to the different settings of the MAC layer parameters that regulate the access phase. We have performed different simulations, varying the N parameter and considering different possible values for the access phase parameters, such as c, T 16, and W min. The results are shown in Figs. 2 and 3 respectively for what concerns the mean packet delay and the mean goodput for a CBR flow. 10 repeated simulations of 400 s have been performed for each point to have reliable results. These different simulations are aimed to determine the best access parameter values for scenarios with different N values; the adopted criterion is to maximize the CBR goodput 2 provided that the mean CBR packet delay is lower than 80 ms. As for the mean CBR packet delay, we can note that it increases with N and T16. Moreover, 2 The mean CBR goodput may be lower than the source bit-rate value due to packet losses caused by buffer overflow.

5 TABLE II SELECTION OF ACCESS PARAMETERS FOR OUR WIMAX SCENARIO. BPSK 1/2 16QAM 1/2 N c T16 W min 1, ms ms (or 10 ms) 7 (or 3) ms ms 7 the mean CBR goodput reduces with lower c values (the contention phase is more congested and more time is needed for the access with the risk of packet losses due to buffer overflow) and with N due to the congestion of resources. Note that in all the cases in Figs. 2 and 3 the aggregated FTP downlink goodput remains practically constant with an evident reduction only in the case of N = 4 due to the congestion of uplink resources and the consequent delay in sending ACK uplink packets. As a conclusion, we can consider Table II where the best settings of the access parameters are proposed (for BPSK 172 and N = 4 we do not achieve the delay requirement for CBR). Finally, we focus on a configuration with N FTP and N CBR flows as before, but where FTP downlink flows are divided into two groups: one group formed by an SS, experiencing bad channel conditions with a given Packet Error Rate (PER) at the IP level, PER = 3 %; the other group is formed by N 1 FTP flows with excellent channel conditions (no errors). This situation is to account for two different channel conditions. The interest is here to investigate the impact that an increase in the number of FTP flows experiencing good channel conditions has on the FTP flow with PER =3% in the presence of a round robin scheduler. The performance results are shown in Fig. 4. We can note that the FTP flow with PER = 3 % is penalized in goodput by the increase in the number of the other FTP flows since these flows are able to exploit the injection rate reduction due to the frequent losses of the FTP flow with PER = 3 %. This behavior is present with both BPSK 1/2 and 16QAM 1/2, even if with 16QAM 1/2 performance is much better for both FTP downlink flows and CBR uplink ones. VII. CONCLUSIONS AND FUTURE WORK This paper provides a preliminary MAC study for a WiMAX scenario suitable for the WEIRD project. In the current NIST simulator, only the BE traffic class is implemented with round robin scheduler (the WiMAX Forum is working to improve the NIST simulator). Then, we have considered a scenario with FTP downlink and video-cbr uplink traffic flows with the aim to select appropriate settings for the access parameters. A further refinement of this work is needed to implement a scheduler that accounts for traffic differentiation and the priority of CBR traffic (UGS class) with respect to other traffic classes (e.g., BE class); moreover, the scheduler should account for fairness issues to compensate for the resource unbalance due to SSs experiencing different channel conditions. Fig. 2. Mean CBR packet delay as a function of N, the number of CBR and the number of FTP flows. Fig. 3. Mean CBR goodput as a function of N, the number of CBR and the number of FTP flows. Fig. 4. N FTP users experiencing different downlink channel conditions. REFERENCES [1] Web site of the WEIRD Project with URL: [2] IEEE , IEEE Standard for Local and Metropolitan Area Networks, Air Interface forfixed Broadband Wireless Access Systems, October [3] Loutfi Nuaymi, WiMAX: Technology for Broadband Wireless Access. John Wiley and Sons, Chichester, England, [4] J. Chen, W. Jiao, H. Wang, A Service Flow Management Strategy for IEEE Broadband Wireless Access Systems in TDD Mode, ICC 2005, pp , May [5] H. Lee, T. Kwon, D.-H. Cho, An Efficient Uplink Scheduling Algorithm for VoIP Services in IEEE BWA Systems, VTC 04, pp , Sept [6] L. Lin, W. Jia, W. Lu, Performanc Analysis of IEEE Multicast and Broadcast Polling based Bandwidth Request, WCNC 2007, pp , [7] Web site with URL: jain/cse574-06/ftp/wimax qos/index.html [8] Web site with URL: [9] WiMAX Forum Web site with URL: