DEWS2008 D8-6 LAN QoS TCP 112 8610 2 1 1 NEC 211 8666 1753 E-mail: emi@oglisochaacjpoguchi@computerorg y-hirano@hyjpneccom t-murase@apjpneccom QoSQuality of Service QoS TCP-AV TCP TCP-AV QoS TCP LAN QoS A Study on Fairness of Bandwidth and Analyze TCP with QoS-guarantee Mechanism on a Wireless Environment Emi ARAI Yumi HIRANO Tutomu MURASE and Masato OGUCHI Ochanomizu University Otsuka 2 1 1Bunkyo-Ku Tokyo 112 8610 Japan System Platforms Research Laboratories NEC Corporation 1753 Shimonumabe Nakahara-ku Kawasaki Kanagawa 211 8666 Japan E-mail: emi@oglisochaacjpoguchi@computerorg y-hirano@hyjpneccom t-murase@apjpneccom Abstract In late years the demand for multimedia communication has been raised extensively To guarantee the QoS is extremely important QoS defined here is to guarantee target bandwidth In order to realize it TCP-AV has been proposed and implemented and it is achieved in a wired environment However wireless environment has problems different from the wired environment When multiple s are communicating it is said that unfairness of the throughput among s is observed caused by control mechanism of the data transmission right As a result there is a possibility that it cannot control the bandwidth effectively Therefore we invested and considered these problems in this paper We have used TCP-AV on wireless environment and evaluated performance of multimedia communication for aiming at better QoS controls by resolving problems Key words TCPwireless LANfairnessQoS 1 TCP/IP QoS VoIP QoS QoSQuality 1 TCP-AV [1] of Service QoS QoS TCP TCP-AV QoS
QoS TCP-AV fair case QoS LAN VoIP throughput of all s TCP-AV 1 TCP 2 1 CSMA/CA CSMA/CACarrier Sense Multiple Access with Collision QoS Avoidance LAN QoS QoS 2 IFS: Inter Frame Space 1 LAN 1 LAN [2] TCP [3] 2 1 1 CSMA/CA 1 LAN APAccess Point TCP-ACK 80211 DIFSDistributed Coordination Function IFS OS CWContention Window AP CW 2 2 LAN 2 1 2 2 80211 LAN TCP-AV DCFDistributed Coordination Function 2 TCP 3 TCP TCP-AV CSMA/CA 4 CSMA/CA 2 LAN TCP MACMedia Access Control TCP 0Mbps 2 2 CSMA/CA TCP 1 are almost equal LAN Terminal ID Unfair case Terminal ID There are huge difference on throughput among s TCP 1 RTT Round Trip Time
MAC IEEE8023 CSMA/CDCSMA with Collision Detection IEEE80211 CSMA/CA 2 CSMA/CD 2 2 1 TCP-ACK ACK (Collision Detection) ACK 1 (Multiple Access) 10 10 1 ACK USB 2 1 1 ACK 2 MAC USB MAC CSMA/CA 10 ACK CSMA/CD 3 USB BUFFALO WLI-UC- AG [4] 2 2 2 BUFFALO WLI3-TX1-G54 [5] BUFFALO WLI- TX4-G54HP [6] AP ACK [3] 10 AP 1 2 54Mbps ACK 100Mbps AP AP ACK AP 10 ACK 1/11 AP ACK ACK ACK 2 4 TCP 2 Receiving TCP ACK <wired> <wireless> Access Point TCP data Transmitting Transmitting Transmitting TCP 2 3 MAC MAC TCP CA CWR TCP CA LOSS 3 (Carrier Sense) 3 MAC TCP-Reno QoS TCP-AV 2 4 1 TCP-Reno TCP-Reno Linux TCP [7] TCP CA OPEN TCP TCP CA RECOVERY ACK SACK Local Device Congestion
4 Local Device 1 Congestion NIC TCP 3 1 AP 3 1 1 LAN AP LAN AP BUFFALO WZR-AMPG300NH [9] 5 ACK 1 UDPUser Datagram SACK Protocol AP [10] [increase of congestion window] TCP_CA_Open normal state duplicate ACK SACK TCP_CA_Recovery abnormal state [reduction of congestion window] 5 AP timeout ECN ICMP source quench Local device congestion TCP_CA_Loss abnormal state TCP_CA_CWR abnormal state 4 2 4 2 TCP-AV TCP-AV TCP 3 1 2 3 TCP/TCP-AV TCP-AV LAN TCP AP AP TCP-ACK TCP 3 2 LAN TCP TCP-AV EBEffective BandwidthNRT Normalized Required ThroughputAVRAchievement Ratio EB LAN IEEE80211g LAN [8] 54Mbps NRT 7 OS 1 18 Linux24 6 7 Linux26 25 25Mbps NRT AVR 35 USB 120 5 7 AP 5 23 23 1 10 Iperf [11] fair-share NRT 5 AP AP 10 AP 300 3 2 1 TCP 6 windows XP 1267
120 20 ACK 05% USB MAC ACK OS USB-type throughput[mbps] 6 LAN ID 3 2 2 10 7 USB 7 USB fair-sharenrt 7 3Mbps fair-share Mbps MAC TCP [12] tcpdump 226Mbps fair-share 504Kbps 2 TCP 8 TCP 3 ACK SACK tcpdump ACK 9 ACK 1 1 ACK 8 9 Converter-type 2 3 3 LAN TCP 3 3 1 TCP [2] 8 9 TCP tcpdump
TCP-AV 25 Iperf 120 10 26 10 6 12 TCP 5 TCP-Reno 10 TCP-AV fair-sharenrt 32%8Mbps 10 3 3 2 4 5 11 12 2 4 5 1 12 Terminal ID2 6 3 4 2 14 3flows NRT NRT100% LAN ACK 25Mbps TCP-AV NRT % 3flows 13 TCP-AV 1 TCP-Reno 2 3 11 fair-share 10% TCP 4 3 4 TCP-AV 3 4 1 3 LAN TCP-AV AVR 13 Flow1 TCP-AV TCP TCP 15 5 TCP-Proxy TCP TCP-AV Flow2 Flow3 13 AVR 9 NRT 40% TCP-Reno 1 2 TCP-AV TCP-Reno 4 5 14 5flows 9 NRT 30% CSMA/CA 3 4 3
AVR(Achievement Ratio)[%] 100 90 80 70 60 50 40 30 20 10 0 14 0 20 40 60 80 100 NRT(NormalizedRequired Throughput)[%] 3 flows( ) TCP-AV- flow TCP-Reno-2 flows (1flows*2s ) 5 flows( ) TCP-AV- flow TCP-Reno-4flows (2flows*2s) AVR TCP-AV 1 3 AP USB TCP tcpdump 2 QoS TCP- AV AVR TCP-AV fair-share 510% QoS UDP TCP-AV 15 AVR 3 4 4 fair-share NRT 20% 5Mbps 16 fair-share AVR 1 LAN fair-share [3] SPilosofRRamjeeYShavittPSinha Understanding TCP fairness over Wireless LAN INFOCOM 2003 1-3 April 2003 [4] BUFFALO AirStation 54Mbps 80211 abg (WLI-UC-AG): TCP-AV fair-share 5% [5] BUFFALO 24GH 54Mbps wireless LAN Ethernet conversion media converter (WLI3-TX1-G54: http://buffalojp/products/catalog/item/w/wli3-tx1-g54/ 6 [6] BUFFALO AirStation High Power wireless LAN Ethernet converter 54Mbps (WLI-TX4-G54HP) : http://buffalojp/products/catalog/item/w/wli-tx4-g54hp/ [7] Linux Kerner ver24 implementation: http://wwwlinuxorg/ [8] IEEE80211 wireless LAN medium access control (MAC)and physical layer(phy) specification1999 [9] BUFFALO wireless LAN BB router AirStation NFINITI GIGA 300Mbps 11n conformity (WZR-AMPG300NH) : http://buffalojp/products/catalog/network/wzr-ampg30nh/ [10] Perl program of inspect the size of buffer: http://x68000q-ednet/ 68user/net/udp-2html [11] Iperf: http://dastnlanrnet/projects/iperf/ [12] : iscsi 16 AVR TCP-AV 2 5 4 QoS 1 TCP AP TCP-AV NEC [1] HShimonishi et al Congestion Control Enhancements for Streaming Media IEICE Tranc on ComnVolE89 B No9pp2280-2291 AVR 100% fair-share NRT Sep 2006 25% AVR 70% NRT 40%[2] Anthony CH NG David Malone Douglas JLeith Experimental Evaluation of TCP Performance and Fairness in an 80211e Testbed ACM SIGCOMM 2005pp17-22 Aug2005 http://buffalojp/products/catalog/network/wli-uc-ag/ Vol J87-D-I No 2 pp 216 231 2004 2