MCCA: A High-Throughput MAC Strategy for Next-Generation WLANs

Transcription

1 1 MCCA: A High-Throughput MAC Strategy for Next-Generation WLANs Seongkwan Kim, Youngsoo Kim, Sunghyun Choi, and Kyunghun Jang School of Electrical Engineering and INMC, Seoul National University, Seoul, , Korea New Radio Access PT, Samsung Advanced Institute of Technology, Yongin, , Korea skim@mwnl.snu.ac.kr, KimYoungsoo@samsung.com, schoi@snu.ac.kr, and khjang@samsung.com Abstract WLAN technology has been shown a revolutionary development during the last decade. Recently popularized IEEE a/gbased products can support up to 54 Mbps (Physical layer) rate and give much freedom to access the Internet wirelessly. However, the MAC (Medium Access Control) protocol has relatively large overhead in order to robustly deal with the unreliable wireless nature, and hence, the throughput performance is much worse than the underlying rate. Moreover, along with many emerging applications and services over WLANs, such as VoWLAN (Voice over WLAN) and audio/video streaming, the demands for faster and higher-capacity WLANs have been growing recently. In this article, we propose a new MAC protocol for the next-generation high-speed WLANs. The proposed MAC, called MCCA (Multi-user polling Controlled Channel Access), is composed of two components: (1) MLA (Multi-Layer frame Aggregation), which performs aggregations at both MAC and ; and (2) MUP (Multi-User Polling) being used to reduce the contention overhead and, in turn, achieve higher network utilization. MCCA is compared with the e EDCA (Enhanced Distributed Channel Access) MAC. Highly enhanced MAC efficiency can be achieved by applying MCCA and we show the improved MAC performance in terms of the aggregate throughput of (non-)qos flows with relevant QoS requirements. Index Terms IEEE WLAN, frame aggregation, multi-user polling, throughput, and QoS I. INTRODUCTION In recent years, IEEE WLAN has gained a prevailing position in the market for the (indoor) broadband wireless access networking [1]. The number of IP applications operating over WLANs, such as VoWLAN (Voice over WLAN), audio/video streaming, and file transfer, has increased, and the required bandwidth for such traffic has continuously augmented as well. In consequence, the commodity devices equipped with the high-speed (Physical layer), i.e., supporting up to 54 Mbps rate with IEEE a/g, have recently become popular to WLAN users. The , however, is still known to have large overhead for MAC (Medium Access Control) and operations restraining the throughput performance much worse compared with the underlying high rate. For example, due to such large overhead, the maximum achievable throughput of the DCF (Distributed Coordination Function) is only about 30 Mbps even with the a/g. In [2], the authors demonstrate that by simply increasing the rate without reducing the addressed overhead, the enhanced throughput is bounded around 100 Mbps, even if the rate goes to infinity. It infers that further effort in terms of the protocol efficiency of the MAC is required. Moreover, demands for higherspeed WLANs have been accelerated along with the evolution of the wired counterparts, e.g., 1 or 10 Gbit Ethernet and FTTH (Fiber to the Home) technologies. The protocol overhead in the WLANs can be classified into two categories: header-oriented and access-controloriented. In the , an (MAC Protocol Data Unit), which is generated by adding a MAC header to an MSDU (MAC Service Data Unit) coming from the higher layer, is the unit of data exchanged between two peer MAC layers using the service of the underlying layer, i.e.,. Accordingly, an becomes a PSDU ( Service Data Unit) at the, and then, a PPDU ( Protocol Data Unit) is generated by adding a preamble/header to the PSDU. Considering the fact that a single MSDU becomes a PPDU after adding the fixed-size MAC header and preamble/header, the appended preamble and headers for each IP datagram become relatively large overhead as the size of IP datagram decreases. The DCF is a CSMA/CA (Carrier-Sense Multiple Access)-based medium access method. Each WLAN keeps backing the medium access off in order to alleviate contention and resolve collision with other s. However, as the number of contending s increases and, in turn, s suffer from more frequent collisions, the overhead induced by the backoff increases. Thanks to the fully controlled medium access in the PCF (Point Coordination Function) [1] and the e HCCA (HCF Hybrid Coordination Function Controlled Channel Access) [5], such backoff overhead is eliminated. However, per- poll transmissions contribute to a different type of overhead in both PCF and HCCA. Furthermore, such polling overhead increases as the number of polled s increases. This article has been accepted to IEEE Wireless Communications: special issue on Medium Access Control Protocols for Wireless LANs that will be issued in Feb Guest editors on this special issue are Mohammad S. Obaidat, Petros Nicopolitidis, Petre Dini, and Jung-Shian Li.

2 2 In this article, we propose a new MAC protocol, MCCA (Multi-user polling Controlled Channel Access), to cope with the problem of the legacy MAC protocols discussed above. MCCA consists of two components, MLA (Multi-Layer frame Aggregation) and MUP (Multi-User Polling). MLA deals with the inefficient header usage in the by employing two types of aggregation techniques at both MAC and. In addition, MLA s aggregation can also reduce the access-controloriented overhead owing to the backoff. Accordingly, the reduced channel wastage directly changes to throughput gain. MUP is a polling-based controlled medium access method, and it makes use of a multi-user poll. As a result, MUP reduces not only the backoff overhead, but the polling overhead. The standardization activity of IEEE TGn (Task Group N) shares our goal of enhancing MAC efficiency for highspeed WLANs. In fact, the core idea of MCCA has been proposed as one candidate of the n standardization. The current n draft includes a modified version of MCCA and the standardization activity is still ongoing [3], [4]. The effectiveness of MCCA has been evaluated based on several realistic wireless network models, which are the guidelines of TGn for performance evaluation. We next briefly describe the e EDCA, which is the baseline MAC of MCCA, and then present the proposed MCCA. After providing simulation results and analysis to support our claim that the proposed MAC results in significantly improved MAC efficiency and QoS (Quality-of-Service) guarantee, we close the article with concluding remarks. II. EDCA: BASELINE MAC IEEE e EDCA is designed to provide differentiated and distributed medium access for frames belonging to eight different user priorities, which are mapped into four ACs (Access Categories) [5]. Accordingly, more favored transmission attempts are stochastically guaranteed for higher-priority frames. An EDCA contends for medium access in the same manner, i.e., CSMA/CA, as a DCF does. Four parameters that prioritize and control the way to grab the medium vary along with mapping ACs: (1) AIFS (Arbitration Interframe Space); the minimum time interval before contention, (2) the minimum, (3) the maximum contention windows (CW min and CW max ), and (4) TXOP (Transmission Opportunity). In an EDCA, four transmit queues are equipped along with four ACs, where EDCA contention algorithm is implemented based on corresponding parameters. In particular, the smaller AIFS, CW min, and CW max, the shorter the medium access delay for the corresponding priority and, in turn, the more capacity share for a given traffic condition. A TXOP, which is a time interval when a particular has the right to initiate frame exchange sequences onto the wireless medium, is defined by a starting time and a maximum duration. In general, the longer TXOP duration is assigned to the higher-priority flow for QoS guarantee. Unlike the ACK (Acknowledgement) scheme, which is a stop-and-wait ARQ (Automatic Repeat request), the e defines a new selective ARQ scheme, called BACK (Block ACK), in order to improve the MAC efficiency. Specifically, during a TXOP, a TXOP-initiator transmits a number of data frames without receiving corresponding ACK frames. After the bursttransmission of data frames, the initiator transmits a control frame, called BAR (Block ACK Request). The receiver of the BAR should respond with a BA (Block ACK). 1 BACK bitmap is used in the BACK to inform which frames are not received correctly. III. MCCA: MULTI-USER POLLING CONTROLLED CHANNEL ACCESS MCCA consists of two components, MLA and MUP. Each of MCCA components contributes toward enhanced MAC efficiency; besides, an incorporative usage of both components can synergistically maximize the aggregate network throughput performance. A. MLA: Multi-Layer frame Aggregation The first component of MCCA, MLA, is composed of two types of hierarchical aggregations. Along with where an aggregation is performed, i.e., either MAC or, aggregation schemes are referred to as A- and A-PSDU, respectively. The basic operational example of MLA is illustrated in Fig. 1. By definition, A- can be applied for a number of s to be aggregated into a single PSDU, and, similarly, A-PSDU generates a PPDU comprising one or more number PSDUs. Compared with the concept of the s PPDU and PSDU generations as addressed above, the MAC efficiency can be drastically ameliorated by MLA. We describe the details of MLA as follows. 1) A-: A- can be achieved with multiple s irrespective of their traffic identities. 2 In other words, a number of s can be aggregated into a PSDU as long as they are destined to the same receiver. In Fig. 1, the first three s, which are destined to DA1, compose a single PSDU by A-. Therefore, based on this example, only a single backoff process is needed for three transmissions. Moreover, preamble and header are needed to be attached only once, instead of three times. No doubt we can achieve enhanced MAC efficiency with A-. 1 There are two types of BACK procedures in the e, which are immediate BACK and delayed BACK. The former is explained here. 2 In this article, we assume that each MSDU is differentiated by its traffic identity, which is determined by its QoS requirement and belonging traffic flow [5].

3 3 In order to better achieve the improved MAC efficiency, we introduce a header compression scheme with two of newly designed MAC header formats in A-. When an MSDU becomes the first in an A- frame, a legacy MAC header is used (referred to as L in Fig. 1). The following MSDUs have different types of MAC headers along with their traffic identities. If MSDUs followed by the first MSDU have the identical traffic identity to the preceding MSDU, a simpler format of MAC header (i.e., omitting all duplicated fields from L) is used, namely, type-1 compressed MAC header (C1 ). In contrast, MSDUs, which are heterogeneous in traffic identity, need to be differentiated from the other (e.g., the starting MSDU in the A- frame). Type-2 compressed MAC header (C2 ), in which a QoS control field is included to specify the different traffic identity, should be used in this case. Compressed headers should be regenerated based on the starting s MAC header (L) by the receiver. With the header compression scheme, the considered A- makes the MAC efficiency even higher, and hence, more improved throughput can be expected. With A-, an ( Delimiter) is attached in front of each aggregated. plays a role to robustly delimit the s within the aggregate by the receiver. Consequently, the effective packet error probability of each aggregated should be identical to that of non-aggregated. 2) A-PSDU: With the second frame aggregation scheme, A-PSDU, multiple PSDUs can be aggregated into a single PPDU without additional preambles and headers, irrespective of where they are destined to and what rate they should be transmitted in. Fig. 1 also illustrates an example of A-PSDU operation when two PSDUs, which are destined to different receivers with different rates, are conveyed by a single PPDU. Similar to the purpose of in A-, a PD (PSDU Delimiter) precedes each PSDU in A-PSDU, and it allows to transmit each PSDU at a different rate according to the desired rate to each destined. A PD should be transmitted at the lowest rate (e.g., 6 Mbps of the a), in order for all the s to be able to decode it. With the help of MCS (Modulation Coding Scheme) information specified in each PD, a number of PSDUs being transmitted at different rates can be aggregated into a PPDU using A-PSDU, and successfully decoded by corresponding receivers. B. MUP: Multi-User Polling As the second component of MCCA, we present a controlled medium access scheme, MUP, based on the e EDCA. In order to maximize MAC efficiency, the considered MUP provides two distinctive features. First, unlike PCF and HCCA, MUP deals with both uplink and downlink traffic streams during the controlled access period. Second, an MUP coordinator (i.e., AP) can send multiple polls to different polled s at the same time. The poll frame dealing with this simultaneous polling to multiple s is named MPoll, and its format is carefully designed to support such features. Lo et al. proposed an analogous multi-user polling method for WLAN [6]. Their approach is mainly focused on supporting QoS traffic by using a multi-user poll under DCF. However, if we use the e EDCA instead of DCF, further QoS satisfaction can be expected. In addition, our MUP is synergistically employed with MLA for the purpose of improving MAC efficiency, which is the major difference from the multi-user polling method in [6]. An MPoll provides the control of flows for downlink and uplink within a specific period, called service period. An MPoll conveys a number of downlink and uplink MAP data structures, where such MAPs handle the exchanges of aggregated frames that are generated by previously described MLA (i.e., A- and A-PSDU). The duration field in an MPoll is set to a long enough value to make non-polled s (including legacy s) set NAV (Network Allocation Vector) not to access the medium during the service period. An MPoll specifies the number of A- frames in downlink and the number of s :DA1 MSDU MSDU MSDU MSDU :DA1 :DA1 :DA2 MAC-layer Aggregation ( A- ) -layer Aggregation ( A-PSDU ) PD PSDU PD PSDU basic rate data rate 1 basic rate data rate 2 Legacy MAC Header type-1 Compressed MAC Header type-2 Compressed MAC Header PD Delimiter PSDU Delimiter Frame Check Sequence Fig. 1. An illustrative example of MLA operation.

4 4 in uplink to indicate how many A- frames are destined to particular s in a particular rate, and how many s will get polls to access the medium, respectively. In order to properly guarantee the QoS requirement of each flow from a, an appropriate TXOP value is assigned to each polled, and it is also specified in every MPoll for each polled. Considering the significant use of MPoll, the lowest transmission rate is desired for the MPoll transmission. Fig. 2 is an illustrative example of how MUP works. An AP can access the channel with higher priority than s by deferring a PIFS (PCF Interframe Space) time interval, which is shorter than DIFS and any other AIFS time intervals. Whenever an AP sends an MPoll after a PIFS time interval, a service period is initiated. When a receives an MPoll, it sets its backoff counter to an appropriate value, which is implicitly assigned according to the polling sequence specified in the MPoll. For example, if the number of polled s is three, the maximum value of backoff counters is two as illustrated in Fig. 2, and the backoff counter is assigned to each as specified in the polling sequence (e.g., from zero to two in Fig. 2). A service period is divided into two sub-periods, downlink and uplink periods. The service-period initiator, AP grabs the medium by sending its frames after a PIFS time interval again. The uplink period starts after DIFS idle time and a whose backoff counter becomes zero obtains an opportunity to send its A- frames during the assigned TXOP duration. Thanks to the NAV setting, which covers the remaining service period, non-polled and legacy s cannot access the medium when the AP and polled s exchange their A- frames. The NAV duration specified in an MPoll covers the whole time intervals of the downlink TXOP limit and all possible uplink TXOP limits. It is possible that a CF-End (Contention-Free End) frame 3 is sent by the AP when there is no more uplink traffic from the polled s. After the completion of a service period, either an EDCA-based medium access or a new MCCA service period starts depending on the network operating policy. During the downlink period, A-PSDU can be applied, and hence, each receives one or multiple A- frames destined to itself at its receiving time, which is calculated based on information conveyed by the preceding MPoll. As illustrated in Fig. 2, during the uplink period, each follows the EDCA channel access rule defined in the e except two facts: First, the backoff counter is not specified by the, but assigned by the AP. Second, during a service period, the minimum idle delay before contention, AIFS is set to DIFS for all polled s since there is no need to prioritize the medium access of polled s under MUP. After counting the maximum backoff counter, which the AP assigned to one of the polled s using an MPoll, AP can be aware of no more transmission from polled s, and then, the AP sends a CF-End frame with PIFS deference to end the service period. By receiving the CF-End, all s including non-polled and legacy s are informed of the end of a service period, irrespective of their NAVs. 3 According to [1], the reception capability of a CF-End frame is a mandatory feature of an based. In other words, a backward compatibility problem in terms of CF-End reception of a legacy should not arise when MUP is employed. TXOP limit AP MPoll A- A- CF- END A-PSDU 1 RX mode 0 A- TXOP limit 2 RX mode legacy NAV service period A- TXOP limit PIFS PIFS DIFS DIFS PIFS AIFS[AC] 1 0 normal PSDU Fig. 2. An illustrative example of MUP operation.

5 5 C. Hidden Station Management While the medium access in MUP is designed to be fully controlled by an AP, the essential access mechanism is based on contention among polled s. Therefore, the hidden problem might also happen in MCCA. To mitigate the potential hidden problem in an MCCA-based WLAN, we consider a hidden management method in this article. When a attempts to join an MCCA-based WLAN, it receives a number of MPolls and learns a polling sequence before associated to the AP. Suppose that the new senses the idle channel, while it is aware of who is the next polled and expected to send its A- frame. Accordingly, the new records the next polled as a hidden from itself. In order for the new to correctly detect the existence of hidden s, the AP rotates the polling sequence for every next service period without loss of QoS guarantee for all associated s. After repeating a series of hidden learning procedures, the new reports its hidden s to the AP when it is finally associated. The AP collects such hidden information from all the s and then, runs a grouping algorithm to establish non-hidden groups. Such a grouping algorithm should generate as small number of groups as possible, while group members are not hidden from each other. The grouping algorithm should be repeatedly initiated for every new association. If a collision happens during a service period, the AP can detect a hidden- pair and then, starts re-grouping s. D. Error Recovery via Block ACK Thanks to the MUP s controlled medium access and the considered hidden management method, frame collisions are not likely to happen during a service period. Transmission errors, however, might occur due to unreliable wireless channel condition. More harmful effect due to channel-error-caused transmission failures can be expected as most of frames are transmitted in a form of aggregation in MCCA. Therefore, the careful design of an error recovery mechanism is highly desirable for MCCA. We employ a fine-tuned BACK scheme based on EDCA BACK, called MCCA BACK, as the error recovery mechanism of MCCA. The basic operational usage of MCCA BACK is identical to that of EDCA BACK: the format of BAR and BA, and the usage of BACK bitmap. In MCCA, however, either BAR or BA is aggregated into an A- frame. Moreover, during the downlink period in MUP where a BAR is added to each of A- frames, no corresponding BA is transmitted from BAR-received polled s, and the BA transmissions are delayed to the following uplink period for the purpose of further enhanced MAC efficiency. Considering the addressed difference of BACK usage, we call the exchanged BAR and BA frames by the names of R (MCCA Block ACK Request) and (MCCA Block ACK), respectively. In addition, an AP should give a poll to a if the is one of the receivers of a downlink A-PSDU frame in order for the AP to correctly receive the corresponding BA. For better understanding of MCCA BACK, an illustrative example is given in Fig. 3. During a downlink period, an R follows each A- frame. The reception of corresponding s from s is delayed to the following uplink period. When a, which received an A- frame during the downlink period, grasps the medium and generates its A- frame, a delayed is aggregated preceding the transmitted A- frame. An R also follows the transmitted data frame. As illustrated in Fig. 3, unsuccessful frame receptions are informed by and recovered by retransmissions. Such retransmissions are also carried by A- with time interval, and a recovery procedure continues by the successful completion of the A- frame transmission. If a TXOP is expired with unacknowledged frames, the or the AP resumes retransmission attempts after the service period, i.e., during an EDCA contention period, or in the next service period. TXOP limit TXOP limit TXOP limit AP MPoll R R R CF-END 1 2 A-PSDU RX mode 0 RX mode PIFS PIFS DIFS R A- 1 0 service period DIFS R R PIFS Fig. 3. An illustrative example of MCCA error recovery procedure.

6 6 IV. EVALUATION We evaluate the effectiveness of MCCA in comparison with EDCA using ns-2 [8]. We have modified the simulator by enhancing the original DCF module to support the e EDCA and the proposed MCCA. EDCA parameters (introduced in Section II) are set to the recommended default values as specified in the e standard. In addition, the a is employed for our evaluation except the rate. Available maximum rate for both MCCA and EDCA MAC protocols is configured as 216 Mbps, which can be achievable based on a MIMO-based high-rate model. In particular, by applying 2 2 MIMO-SDM (Multiple-Input Multiple-Output Spatial Division Multiplexing) and multiple channel bonding technique (i.e., two a channels (of 20 MHz) are combined together, and hence, the bonded channel of 40 MHz can be used for the communications), at most four times faster rate can be achieved. Each selects its rate among 9 different rates (from 13.5 to 216 Mbps) by looking up the predetermined rate selection table. The rate selection table has been determined in advance with consideration of path loss between the transmitter and the receiver based on the channel model. Therefore, the selected rate by looking up the table becomes the best rate, so that we observe the maximum capacity achieved by considered MAC protocols. Moreover, we assume that all s are fully controlled by an AP in a service period. Accordingly, neither transmission failure due to channel error nor frame collision is considered. The superiority of MCCA to EDCA is demonstrated under three different usage models; residential, large enterprise, and hot-spot. Each of usage models represents credible worst-case mixtures of applications in a realistic environment as inferred by its name. The combination of considered applications, which consist of VoIP, video phone/conferencing, streaming audio/video, video gaming, and unidirectional FTP (representing a file transfer) flows, makes up each of usage models. For example, hot-spot model is designed by assuming that, on average, 15 VoIP devices, 9 audio/video streaming applications, and 10 FTP flows exist in such an environment. We make use of realistic QoS requirements in terms of delay bound for considered QoS-constraint applications. In addition, an appropriate AC is matched to each application, considering the delay bound characteristics. Detailed characteristics of considered applications are described in Table I. We compare the aggregate throughput performance of QoS and non-qos flows based on EDCA and MCCA for each usage model. In particular, when we deal with QoS applications, only the frames that satisfy their corresponding QoS requirements (delay bounds as described in Table I) are collected to be used for throughput measurement. In other words, if the achieved throughput does not reach the maximum achievable throughput (i.e., offered load), in a specific usage model, it means that the required QoS could not be fully guaranteed with the considered MAC protocol. Fig. 4 represents the aggregate throughput performance of EDCA and MCCA in three different usage models. Six bars on the left-hand side represent the aggregate throughput results of QoS flows in considered usage models, and the right-hand-side bars stand for those of non-qos flows. The illustrated thick solid line represents the aggregate offered load of all QoS flows in a specific usage model (corresponding values are specified above the lines). For example, the aggregate offered load of QoS flows in the hot-spot model is Mbps, in which the offered load of 15 VoIP and 9 audio/video streaming flows are all cumulated. In residential and large enterprise models, we observe that both EDCA and MCCA well achieve the required delay bound for all QoS flows. However, MCCA s enhanced MAC efficiency enables more non-qos flows to be allowed, improving the throughput by 2202 % and 646 % in residential and large enterprise models, respectively. The rationales of the highly enhanced throughput of non-qos flows are two-folds. First, the overhead induced by protocol headers and backoff interval is much reduced by MLA. Therefore, MAC efficiency is improved, and, in turn, more aggregate throughput can be achieved. The second reason comes from the burst-transmission characteristics of TCP, which is employed as the transport layer protocol for all non-qos flows. Although we found that MLA enables more non-qos flows to enter the network, AP s downlink queue for non-qos flows is fully occupied in all cases when only MLA is used. It means that the downlink queue of the AP becomes a bottleneck for non-qos flows. If such congested frames can be served faster, TCP senders in both uplink and downlink directions will get more TCP ACKs, and hence, more TCP segments will be sent up to the TCP receive window size. MUP s aggregate downlink transmission (i.e., A-PSDU) can resolve the bottleneck at AP s downlink queue, and, in consequence, much higher aggregate throughput can be obtained by using both MLA and MUP. Moreover, MUP TABLE I APPLICATION CHARACTERISTICS Application Delay bound (msec) Offered load (Mbps) VoIP 30 msec QoS Video phone/conferencing 100 msec flows A/V streaming 200 msec Video gaming 50 msec 1 Non-QoS Internet file transfer N/A N/A flows Local file transfer N/A N/A

7 7 further reduces the access-control-oriented overhead thanks to its multi-user polling-based medium access. We also observe that, with EDCA, the required delay bound of all QoS flows cannot be fully guaranteed in the hot-spot model. Only about 58 % flows out of all aggregate offered load of QoS flows satisfy their delay-bound requirements. Moreover, EDCA achieves only 0.11 Mbps aggregate throughput for non-qos flows. The major reason for this undesirable performance of both QoS and non-qos flows is the intensified contention due to relatively larger number of active s than in other usage models. Accordingly, the collision probability in the hot-spot model becomes elevated, and hence, most of the flows suffer from low service quality. In contrast, MCCA perfectly achieves the QoS requirements of all QoS flows in the same environment thanks to the controlled channel access and eliminated collision loss. Moreover, Mbps aggregate throughput achievement is observed for non-qos flows, which is a drastic improvement compared with that of EDCA. V. CONCLUDING REMARKS In this article, we present a new high-throughput MAC protocol, called MCCA, for the next-generation high-speed WLANs. MCCA is composed of two components, namely, MLA and MUP. MLA performs aggregation at both MAC and, and hence, can aggregate frames with different QoS requirements and destined to different receivers. A-PSDU, however, has the nature that the aggregated frame should be well scheduled in order for receiving s to recognize it easily and correctly. Accordingly, we introduce MUP for the purpose of scheduling the A-PSDU frame transmissions. Moreover, by employing MPoll, we achieve even higher MAC efficiency thanks to the usage of the minimum-size backoff counters, which are uniquely assigned by the AP. Simulation results show that MCCA outperforms EDCA in terms of the aggregate throughput of non-qos flows without any sacrifice of the required delay bounds of QoS flows. In addition, we observe that MCCA can contribute toward guaranteeing QoS requirements in a heavily contending environment (i.e., hot-spot model), where EDCA guarantees delay bounds only for 58 % of QoS flows. ACKNOWLEDGMENT This work was supported in part by the Korea Science and Engineering Foundation (KOSEF) grant funded by Korea Government (MOST) and the Korea Research Foundation grant funded by Korea Government (MOEHRD) (R ). REFERENCES [1] IEEE , Part 11: Wireless LAN Medium Access Control (MAC) and Physical Layer () specifications, IEEE Std , Aug [2] Y. Xiao and J. Rosdahl, Throughput and Delay Limits of IEEE , IEEE Commun. Lett., vol. 6, no. 8, pp , Aug [3] IEEE Working Group Homepage: [4] IEEE n, Part 11: Wireless LAN Medium Access Control (MAC) and Physical Layer () specifications: Enhancements for Higher Throughput, Draft supplement to IEEE Std , Draft 2.0, Feb [5] IEEE e, Part 11: Wireless LAN Medium Access Control (MAC) and Physical Layer () specifications: Medium Access Control (MAC) Quality of Service Enhancements, Supplement to IEEE Std , Nov [6] S.-C. Lo, G. Lee, and W.-T. Chen, An Efficient Multipolling Mechanism for IEEE Wireless LANs, IEEE Trans. Comput., vol. 52, no. 6, pp , June [7] IEEE , Part 11: Wireless LAN Medium Access Control (MAC) and Physical Layer () specifications, IEEE Std , Jun [8] The Network Simulator ns-2.: Seongkwan Kim received the B.S. and M.S. degrees in electronics engineering from Korea University in 2002 and 2004, respectively. He worked as a visiting scholar at Hewlett-Packard Laboratories, Palo Alto, CA, USA in He is currently pursuing the Ph.D. degree at the School of Electrical Engineering and Computer Science, Seoul National University, Seoul, Korea. His current research interests include IEEE based WLAN and wireless mesh networking with the emphasis upon transmission rate control algorithm, design of enhanced MAC protocol, design of wireless mesh routing metric/protocol. He is a winner of the 12 th Samsung Humantech Thesis Prize from Samsung Electronics in He is a student member of the IEEE since skim@mwnl.snu.ac.kr

8 8 Youngsoo Kim received the B.S. degree in information and engineering from Korea Univeristy in 1997, and received the M.S. degree in computer and information science from the Ohio-State University in He is currently pursuing the Ph.D. degree at the School of Electrical Engineering and Computer Science, Seoul National University. In 2000, he joined Samsung Advanced Institute of Technology, and is currently working as a senior research member. His research interests include Medium Access Control (MAC), cognitive radio MAC technology, power control, and resource management of next-generation wireless access. KimYoungsoo@samsung.com Sunghyun Choi is currently an associate professor at the School of Electrical Engineering, Seoul National University (SNU), Seoul, Korea. Before joining SNU in September 2002, he was with Philips Research USA, Briarcliff Manor, New York, USA as a Senior Member Research Staff and a project leader for three years. He received his B.S. (summa cum laude) and M.S. degrees in electrical engineering from Korea Advanced Institute of Science and Technology (KAIST) in 1992 and 1994, respectively, and received Ph.D. at the Department of Electrical Engineering and Computer Science, The University of Michigan, Ann Arbor in September, His current research interests are in the area of wireless/mobile networks with emphasis on wireless LAN/MAN/PAN, nextgeneration mobile networks, mesh networks, cognitive radios, resource management, data link layer protocols, and cross-layer approaches. He authored/coauthored over 100 technical papers and book chapters in the areas of wireless/mobile networks and communications. He holds 11 US patents, five European patents, and six Korea patents, and has tens of patents pending. He is serving or served as a General Co-Chair of COMSWARE 2008, and a Technical Program Committee Co-Chair of ACM Multimedia 2007, IEEE WoWMoM 2007 and IEEE/Create-Net COMSWARE He was a Co-Chair of Cross-Layer Designs and Protocols Symposium in IWCMC 2006, the workshop co-chair of WILLOPAN 2006, the General Chair of ACM WMASH 2005, and a Technical Program Co-Chair for ACM WMASH He is currently serving and has served on program and organization committees of numerous leading wireless and networking conferences including IEEE INFOCOM, IEEE SECON, IEEE MASS, and IEEE WoWMoM. He is also serving on the editorial boards of IEEE Transactions on Mobile Computing, ACM SIGMOBILE Mobile Computing and Communications Review (MC2R), and Journal of Communications and Networks (JCN). He is serving and has served as a guest editor for Pervasive and Mobile Computing (PMC), ACM Wireless Networks (WINET), IEEE Journal on Selected Areas in Communications (JSAC), IEEE Wireless Communications, Wireless Personal Communications, and Wireless Communications and Mobile Computing (WCMC). He gave a tutorial on IEEE in ACM MobiCom 2004 and IEEE ICC Since year 2000, he is an active participant and contributor of IEEE WLAN Working Group. He is the recipient of the 2005 Best Teacher Award from the College of Engineering, Seoul National University. From 2004 to 2007, his research on IEEE (e) WLAN QoS is supported by Ministry of Science and Technology (MoST), Korea under Young Scientist Award program. Dr. Choi was a recipient of the Korea Foundation for Advanced Studies (KFAS) Scholarship and the Korean Government Overseas Scholarship during and , respectively. He also received a Bronze Prize at Samsung Humantech Paper Contest in He is a senior member of IEEE, and a member of ACM, KICS, IEEK, KIISE. schoi@snu.ac.kr KyungHun Jang received the BS, MS, and PhD degrees in electronics engineering from Korea University, in 1993, 1995, and 1998, respectively. He was an assistant professor at the Research Institute for Information and Communication, Korea University, in In 1999, he joined Corporate R&D Center, Samsung Electronics as a senior engineer. Since 2001, he has been with Samsung Advanced Institute of Technology as a principal engineer. His research interests include cognitive radio, Medium Access Control, wireless LAN/PAN and next generation wireless systems. khjang@samsung.com

9 Residential: EDCA Residential: MCCA Large enterprise: EDCA Large enterprise: MCCA Hot-spot: EDCA Hot-spot: MCCA Aggregate offered load Aggregate throughput (Mbps) Residential Hot-spot Large enterprise QoS flow 1.19 non-qos flow 0.11 Fig. 4. Aggregate throughput comparison between EDCA and MCCA in three usage models.