QoS issues in Wi-Fi-WMM based triple play home networks Yun Tao Shi Jean-Marie Bonnin Gilles Straub Thomson, France INRIA/IRISA, France Thomson, France yun-tao.shi@thomson.net jm.bonnin@enst-bretagne.fr gilles.straub@thomson.net Abstract In this paper, we address QoS (Quality of Service) issues of video transmission over WiFi-WMM based triple play home networks. Based on NS (Network Simulator version) tool, a simulation study was conducted using triple traffic patterns (video, voice and data) under various home network scenarios. The results basically illustrate that WMM (Wi-Fi Multimedia) cannot provide sufficient QoS mechanisms to guarantee quasi error free video delivery in a home network when video is competing with voice and data.. Introduction IEEE. standard, also known as the commercial trademark Wi-Fi, is being widely accepted. It is used not only in businesses and hot-spot public spaces, but also in residences. For example, it will be attractive to deliver digital TV programs to any kind of display, from a TV set, a PC or a laptop to a PDA everywhere inside a home without the need for additional wires. The first obstacle that appears when using Wi-Fi in triple play home networks is that it lacks QoS mechanisms necessary for video and voice applications. To enhance the MAC (Medium Access Control) level QoS, WMM is proposed by the Wi-Fi Alliance as a profile of the upcoming.e standard to improve the original. multimedia transmission capabilities []. Some studies have been reported to prove the effectiveness of MAC service differentiation provided by the.e EDCF (Enhanced Distribution Channel Function) Compared to the DCF (Distribution Channel Function) [] [] []. In our study, we are focusing on evaluating the capability of WMM to satisfy the high QoS requirement of video transmission in competition with voice and data in home networks. Retry limit is a station-based tunable parameter. In our simulation, we will also explore the impact of retry limit to video delivery in Wi-Fi-WMM based home networks. This paper is organized as follows. In Section, we introduce a multicast to unicast conversion mechanism which is the first step to enhance the QoS for video delivery over Wi-Fi home networks. In Section, we briefly introduce the background and information about WMM. Section describes the simulation setup for evaluating the performance of WMM in home networks using NS simulator tool. In Section, simulation results and analysis are reported. Final thoughts are given in Section.. Multicast to unicast Live TV usually relies on IP (Internet Protocol) multicast. However,. wireless channels are prone to location-dependent errors, meaning that each client is likely to lose a different set of video packets from the video stream. As a result, IP multicast over. links proves to be inadequate due to the lack of acknowledgement mechanism. Unicast transmissions over. wireless networks are more reliable than multicast. Each packet is protected through CRC and is acknowledged by the receiver, then, corrupted packets can be retransmitted. Thus, transforming multicast packets to unicast packets at the AP side is a first step towards enhancing the QoS for video delivery over Wi-Fi networks. Multicast IP video delivery uses IGMP (Internet Group Management Protocol) for IP address selection. Multicast to unicast conversion can be done at layer by snooping multicast IP addresses and source MAC addresses contained in IGMP queries. Once the AP (Access Point) knows which station has registered to which multicast address, it can operate a layer address translation (multicast to unicast). The limitation of this mechanism is that it can only be applied to a network that contains a limited number of devices, such as in a home network. The reason is that each multicast-to-unicast packet may be transmitted several times if more than one station
requests the same multicast group which increases the necessary wireless bandwidth.. Overview of WMM The legacy. MAC protocol employs DCF to coordinate shared medium access. The DCF is the basis of standard CSMA/CA (Carrier Sense Multiple Access with Collision Avoidance) mechanism with binary exponential backoff. In. protocol, carrier sensing is performed using both a physical mediumdependent method and the virtual carrier-sensing method. All data traffic is treated in a first come and first serve, best effort manner. So, there is no differentiation service to support multimedia application with QoS requirements. To provide QoS features and multimedia support to the., WMM is proposed by the Wi-Fi Alliance as a profile of the upcoming.e standard. WMM is based on EDCF; it defines four AC (Access Categories) for specific types of traffic (voice, video, best effort, and background). Each AC uses different AIFSN (Arbitrary Inter Frame Space Number) and CWmin/CWmax values instead of treating all the traffic with a single DIFS (Distributed Interframe Space) and CWmin/CWmax in DCF. The highest priority is assigned to the AC with the smallest AIFSN, Cwmin/Cwmax values. However WMM does not guarantee that the highest priority flow will be served first. This is due to the fact that the contention window of different AC overlaps and all start at. WMM only increases the probability for high priority flows to be served first. The simulation described below will investigate whether WMM QoS is sufficient for video delivery or not.. Simulation setup NS is a discrete event simulator for networking research []. It provides substantial support for simulation over wired and wireless networks. In this study, all the simulations are performed under NS-. and the patch developed by the Technical University of Berlin that supports.e protocol []. However, this model did not allow running simulations with WMM and legacy devices. Therefore the model has been enhanced to allow different WMM parameters (AIFSN, CWmin, CWmax) for different stations. This allowed selecting WMM parameters for wireless stations close to legacy. settings. Also, the enhanced model allowed selecting different WMM parameters for AP and stations. In our simulation, AC is configured to behave as a legacy. station so STA_C plays the role of a non WMM station. The physical layer is set to IEEE.a, Mbps data transmission rate. AC parameters used in the simulation for AP and wireless station are shown in Table and Table respectively. Table. WMM parameters for AP AC AIFSN CWmin CWmax Table. WMM parameters for STA AC AIFSN CWmin CWmax In our simulation, we assume no hidden terminals and no mobility in the system. If not stated otherwise, fragmentation and RTS/CTS are not used... Simulation topology In the simulation, we focus on infrastructure BSS (Basic Service Set), all the traffics are handled by a single AP; no ad-hoc operation is simulated. A simulation topology with source nodes (S, S), router node (R), four stations and one access point (AP) is shown in Figure. In the simulation, transmission distances between AP and stations are chosen in such a way that they are not hidden from each other. S S.. Error Model node wireless link link R The error model is an important part of performance evaluation in simulations. Usually, simpler models give poorer quality performance predictions. In order to add an error model to wireless networks, each AP.a PHY _D STA_ C Figure. Simulation topology _A _B
wireless node can insert a given statistical error model in either outgoing or incoming wireless channels. In our simulation, we used the multi-stated error model proposed by NS to simulate the wireless channel as a two-state error model. It assumes a good channel state and a bad channel state. Within each state, errors occur according to the independent model with a certain error rate. Conceptually, the next channel state is selected using the transition state matrix. Transitions to the next state ( good or bad ) can occur after the end of the duration of the current state. The parameters of the error model in the simulation are shown in Table. Table. Parameters of the error model bad error rate (packets) % good error rate bad duration ms/ms/ms good duration ms time-based {.,.} transmission matrix {.,.}.. Traffic patterns An Exponential ON/OFF model is used to simulate the voice traffic over IP (we assume the G. codec is used kbps coding rate, ms frame interval, bytes RTP payload size, frames/s,.% talk time), the parameters of the Exponential ON/OFF model are shown in Table. Table. Parameters of voice traffic AC average on duration.ms average off duration.ms bytes kbps CBR (Constant Bit Rate) source over UDP (User Datagram Protocol) Agent model is used to simulate the video stream. The parameters of the CBR video stream are shown in Table. Table. Parameters of video traffic AC bytes Mbps An FTP/TCP (File Transfer Protocol/ Transmission Control Protocol) application model with a of bytes and AC is used to simulate the data traffic. In NS, TCP agents include a set of TCP senders and receivers. Various senders (Tahoe, Reno, NewReno, Sack, Vegas, Fack) obey different congestion and error control schemes. In the simulation, Tahoe TCP agent will be used as the base TCP sender. An Exponential ON/OFF model is used to simulate the Burst UDP traffic. The parameters of the Exponential ON/OFF model are shown in Table. Table. Parameters of data traffic AC average on duration ms average off duration ms bytes Mbps.. s By considering typical multi-service home networks, the simulations were conducted using the following six scenarios: : a baseline traffic scenario that consists of only one unidirectional video traffic (from AP to _A). : same as with the addition of data traffic (i.e., from STA_C to AP). UDP and FTP (upload files) burst are simulated respectively. : same as with the addition of one pair of voice connections (i.e., from AP to _B and vice versa) with the highest priority (AC number ) so as to evaluate the impact of increasing voice traffic. : two unidirectional video traffics (i.e., from AP to _A and AP to _B) with the same AC number. : same as with the addition of data traffic (UDP, FTP (upload files) bursts). : same as with the addition of one pair of voice connections.. Simulation results and analysis.. Retry limit Figure plots the video PRR (Packet Retransmission Rate), video PLR (Packet Loss Rate) and video average MAC end to end delay versus different scenarios for different retry limits (,, and ) under different error model parameters (bad state duration = ms/ms/ms). In Figure, we can see that increasing the retry limit reduces the video packet loss rate, but doesn t increase the MAC overhead. Under different physical
,%,%,%,%,%,%,%,%,%,%,%,%,%,%,%,%,%,%,%,%,% MAC retransmission rate,%,% (a). Bad duration ms Mac packet retransmission rate,%,% ftp udp (b). Bad duration ms Mac packet retransmission rate,%,% ftp udp (c). Bad duration ms Figure. Video PRR, PLR and average MAC delay for different retry limit (,, and ) layer channel models (bad duration ms/ms/ms), it seems that modifying the retry limit has no strong impact on the video MAC retransmission rate. However, increasing the retry limit increases the average MAC end to end delay, especially when video is competing with voice and data... Packet loss In, there is only a video traffic from AP to a wireless station, and no other station contends the wireless medium, so no collision occurs. The reason of packet loss is the packet errors generated by the error model. When the bad duration decreases from ms to ms, the video packet loss rate of reaches zero. But we can also see some packet drops occur when voice and data are added in other scenarios. The packet loss rate for different scenarios (e.g. retry = ) under different error model parameters (bad state duration = ms/ms/ms) is below.%. This result shows that the video lost rate is relatively low using WMM, but it still cannot provide sufficient QoS mechanisms to guarantee quasi error free video delivery in home networks. This is due to the fact that the contention windows of different AC overlap and all start at, WMM only increases the probability of high priority flows to gain TXOP (Opportunity to Transmit), and it cannot guarantee the highest priority flow will be served first. We can observe that packets are mostly dropped in. By analyzing the simulation trace files, we find that queues will build up at AP buffer when the video throughput drops, this ultimately leads to the queue full, and packet drops.. Conclusion In this paper, we try to evaluate whether WMM can provide sufficient QoS mechanisms to guarantee video delivery in home networks, where the video, voice and data transmissions are supposed to be prevailing applications. Our simulation results indicate the following conclusions.
Increasing the MAC retry limit reduces video packets losses and increases the MAC end to end delay. However, it doesn t increase the MAC overhead and has very little impact on the MAC retry rate. Under good wireless channel conditions, WMM mechanism can guarantee the throughput of video and limit the packet loss rate. Although WMM can provide a certain level of service differentiation among multi-service, it still cannot guarantee the high QoS requirements of video delivery in home networks. Wireless physical layer errors are the major source of packet drops, if the wireless channel conditions worsen, even WMM cannot ensure the video transmission without packets getting dropped. The packet drop rate is relatively low and shows that a relatively low FEC could solve remaining issues.. References [] Wi-Fi Alliance, Wi-Fi CERTIFIED for WMM-support for Multimedia applications with QoS in Wi-Fi Networks [] S. Mangold, S.Choi, P.May, O.Klein, G.Hiertw and L.Stibor, IEEE.e Wireless LAN for Quality of Service in Proc. Eur. Wireless, Florence, Italy, Feb., pp. -. [] D.He and C.Q.Shen, Study of IEEE.e EDCF, in Proc. IEEE VTC-Spring, Jeju, Korea, /. [] J.W. Robinson and T.S. Randhawa, Saturation throughput analysis of IEEE.e Enhanced Distributed Coordination Function, IEEE Journal on Selected Areas in Communications, VOL., No., June, pp.-. [] http://www.isi.edu/nsnam/ns/ []http://www.tkn.tu-berlin.de/research/.e_ns/