RealMedia Streaming Performance on an IEEE 802.11b Wireless LAN T. Huang and C. Williamson Proceedings of IASTED Wireless and Optical Communications (WOC) Conference Banff, AB, Canada, July 2002 Presented by Feng Li lif@cs.wpi.edu CS577 Spring 2005 1
Introduction Three fast-growing Internet technologies World-Wide Web TCP/IP to the masses Multimedia streaming real-time, on-demand audio/video to the home Wireless networks freedom from physical constraints of wires (anything, anytime, anywhere) All available and relative low cost This paper explores the convergence of the 3 Focus on Real Media (popular) Focus on IEEE 802.11b (popular) 5/25/2005 CS577 Spring 2005 2
Objectives Characterize network traffic by Real Media Useful for capacity planning Useful for building simulations/models Relationship between wireless channel (error rate, delay, etc) and user quality Use wireless sniffer, correlate with application Ascertain impact of streaming on competing (ie- TCP) traffic Impact of streaming on Internet traffic of interest 5/25/2005 CS577 Spring 2005 3
Outline Introduction Background Methodology Results Related Work Conclusions (done) 5/25/2005 CS577 Spring 2005 4
IEEE 802.11b Wireless LAN (1 of 2) High speed (up to 11 Mbps, 11g up to 54) Specifies physical layer and MAC layer Physical layer allows 1, 2, 5.5, 11 Mbps Higher rates achieved by using sophisticated modulation Header transmitted at 1 Mbps with clocking information (so payload can be transmitted faster) Physical layer has loss, fading and interference Result in corrupted packets, especially at high rates So, dynamically adjust rates based on channel error rate 5/25/2005 CS577 Spring 2005 5
IEEE 802.11b Wireless LAN (2 of 2) Is shared broadcast, so MAC layer regulates access Carrier-Sense Multiple Access with Collision Avoidance (CSMA/CA) or Distributed Coordination Function (DCF) If station wants to send, senses channel If idle for frame time, send packet Otherwise, wait until idle + another frame time + random (double random time) Data sent requires ACK. No ACK, then resend. Give up after 4 tries. Receiver ignores if CRC error. Can be Infrastructure mode (AP) or ad-hoc mode (peer-to-peer) 5/25/2005 CS577 Spring 2005 6
Real Networks Streaming Media (1 of 2) RTSP Server Data: TCP or UDP Buffering Sure Stream Scalable Video Technology Repair 5/25/2005 CS577 Spring 2005 7
Real Networks Streaming Media (1 of 2) Codec, server, client Reliable or unreliable Live or on-demand Header identifies Key frames, decide to retransmit Streaming rate RTSP for communication Control in TCP, data UDP Parameters during session 5/25/2005 CS577 Spring 2005 8
Outline Introduction (done) Background Methodology Results Related Work Conclusions (done) 5/25/2005 CS577 Spring 2005 9
Experimental Environment (1 of 2) Real Server 8.0, Linux, 1.8 GHz P-4, 10 Mbps NIC RealPlayer 8.0, 800 MHz P-3, Cisco Aironet 350 NIC AP lucent RG-1000 WAP, Retransmit limit set to 4 5/25/2005 CS577 Spring 2005 10
Experimental Environment (2 of 2) Video of a rock concert Target rate about 200 kbps above modem, below broadband Short clip 5/25/2005 CS577 Spring 2005 11
Experimental Design Streaming with and without TCP/IP traffic Classify wireless Based on OS status meter TCP background generated from server to client Three traces per experiment Trace at server using tcpdump Trace close to AP using sniffer Trace at client using tcpdump Get wireless and higher layers 5/25/2005 CS577 Spring 2005 12
Outline Introduction (done) Background Methodology Results Related Work Conclusions (done) (done) 5/25/2005 CS577 Spring 2005 13
Baseline Throughput Results Use netperf for 60-seconds, 84 KB receive socket buffer, 8 times Weaker signal, lower throughput Maximum observed, 4.6 Mbps, less than 11 10 Mbps Ethernet not bottleneck Only Poor has too low a throughput 5/25/2005 CS577 Spring 2005 14
Subjective Assessment Playback very smooth for Excellent and Good * For Fair, playback was jerky (lost frames?), but visual quality was good Audio was good for Fair-Excellent For Poor, playback was jerky, some pictures blurry and truncated, audio deteriorated In some cases, setup failed 5/25/2005 CS577 Spring 2005 15
Effect of Wireless Channel (1 of 2) 5/25/2005 CS577 Spring 2005 16
Effect of Wireless Channel (2 of 2) Bursty loss Still residual errors - App has different view of channel - Mostly, expects to be static 5/25/2005 CS577 Spring 2005 17
Application Layer Streaming Rate (1 of 2) Initial phase (10-20 sec) is higher rate (about 3x) Audio always meets target rate (Real favors audio) Excellent and Good similar, meet target video Fair and Poor well below target rate - 17.5 kbps, 12.1 kbps 5/25/2005 CS577 Spring 2005 18
Application Layer Streaming Rate (2 of 2) Excellent and Good similar, meet target video Fair and Poor well below target rate - 17.5 kbps, 12.1 kbps 5/25/2005 CS577 Spring 2005 19
Application-Layer Retransmission NACK based approach reasonable for lost packets Excellent does not lose any Raw loss: - Good has 0.3% - Fair has 10% - Poor has 30% Effective loss: - Excellent and Good have none - Fair has 0.2% audio, 1.3% video (it looked good) - Poor had 7% audio, 28% video (deteriorating) 5/25/2005 CS577 Spring 2005 20
Is That True? One statement: In our experiment, the only packets that miss the deadline are retransmitted packets. page 6, left column. So I doubt this statements: Because some retransmitted packets may meet the deadline. I think the number of retransmitted packets should be greater than what they listed in their paper. 5/25/2005 CS577 Spring 2005 21
Streaming with Competing Traffic Excellent channel 10, 20, 30,40, 50 competing bulk-tcp Should be 460, 230, 150, 115, 92 kbps Asks for more than fair share so not TCP-Friendly 5/25/2005 CS577 Spring 2005 22
Outline Introduction (done) Background Methodology Results Related Work Conclusions (done) (done) (done) 5/25/2005 CS577 Spring 2005 23
Related Work No wireless streaming ( To the best of our knowledge ) Mena et al RealAudio [11] Non-TCP friendly, periodic Wang et al RealVideo [19] Average 10 fps, little full-motion video Loguinov et al MPEG-4 emulation [10] Modem, jitter, asymmetry Chesire at al University workload (Levy) 5/25/2005 CS577 Spring 2005 24
Conclusions Wireless channel has bursty loss but MAC layer retransmission can hide Application layer takes care of most of rest Good and Excellent fine for some streaming Fair and Poor have degraded quality With TCP traffic, RealPlayer not fair 5/25/2005 CS577 Spring 2005 25
Discussion: Shortcomings of their experiments? Subjective Assessment of Streaming Quality. Qualitative Characterization of wireless conditions, based on the Link Status Meter on the Cisco Aironet 350 devices. (eyeball tests)? 68 secs video and low encoding bitrate.. However, in figure 5. From figure 5, the play back duration should be greater than 90 secs with poor signal strength. So I am asking one experiment is enough? ( variability in throughput, and scaling?) 5/25/2005 CS577 Spring 2005 26
Future Work? 5/25/2005 CS577 Spring 2005 27
Future Work Larger scale study (more videos, encodings, ) Effects of mobility Effects on other users on WAP Fragmentation to reduce loss Other technologies (WSM ) Estimating capacity 5/25/2005 CS577 Spring 2005 28
References Mark Claypool, slides for CS529 http://www.cs.wpi.edu/~cs529/f04/slides/kw02.ppt 5/25/2005 CS577 Spring 2005 29