Media Streaming Performance in a Portable Wireless Classroom Network Presenter: Jean Cao Supervisor: Carey Williamson TRLabs & Dept. of Computer Science UoC April 14, 2005
Motivations Media Traffic Characterization Classroom Measurements Extended Experiments: Capacity Bottleneck The “Bad Apple” Phenomenon Conclusions Current and Future Work Outline
Motivations On-Line Education A French Cinematography Class Test Feasible? Performance Issues? Note: Citation numbers are referred to the corresponding references of the paper. Server Client Sniffer Protocol Efficiency? Wireless LAN Portable Wireless Classroom Network [7, 8, 11] Multimedia Streaming [1, 2, 3, 4]
Motivations On-Line Education A French Cinematography Class Test Feasible? Performance Issues? Note: Citation numbers are referred to the corresponding references of the paper. Sniffer Protocol Efficiency? Wireless LAN Portable Wireless Classroom Network [7, 8, 11] Multimedia Streaming [1, 2, 3, 4] Server Client
Motivations On-Line Education A French Cinematography Class Test Feasible? Performance Issues? Note: Citation numbers are referred to the corresponding references of the paper. Sniffer Protocol Efficiency? Wireless LAN Portable Wireless Classroom Network [7, 8, 11] Server Client Multimedia Streaming [1, 2, 3, 4]
Motivations On-Line Education Feasible? Performance Issues? Note: Citation numbers are referred to the corresponding references of the paper. Sniffer Protocol Efficiency? Wireless LAN Portable Wireless Classroom Network [7, 8, 11] Server Client Multimedia Streaming [1, 2, 3, 4] A French Cinematography Class Test
Motivations On-Line Education Note: Citation numbers are referred to the corresponding references of the paper. Sniffer Wireless LAN Portable Wireless Classroom Network [7, 8, 11] Server Client Multimedia Streaming [1, 2, 3, 4] A French Cinematography Class Test Feasible? Performance Issues? Protocol Efficiency?
Motivations On-Line Education Note: Citation numbers are referred to the corresponding references of the paper. Wireless LAN Portable Wireless Classroom Network [7, 8, 11] Server Client Multimedia Streaming [1, 2, 3, 4] A French Cinematography Class Test Feasible? Performance Issues? Protocol Efficiency? Sniffer
Media Traffic Characteristics Digitized 8 min clip from movie “Au Revoir Les Enfants (1998)” Converted to MPEG kbps and 43 fps 400 kbps and 30 fps Total 14,854 video frames Mean frame size 1,641 bytes 90% frames are [1 KB, 2 KB] 1.4% > 4 KB
Media Traffic Characteristics Video Source Traffic Profile at Different Granularities: (a) 1 second(b) 3 seconds(c) 30 seconds(d) 1 minute Constant Bit coarse-grain time scale Variable Bit finer-grain time scale
Classroom Measurements French cinematography class System setup IEEE802.11b Ad Hoc Network (11Mbps theoretical bandwidth) 8 laptops as Clients (mpeg4player) 1 laptop as Server (Apache http server and Darwin Streaming server) 1 laptop as channel monitor (Sniffer software) Procedure: Watch the 8 minute movie clip Complete the Web based multiple-choice quiz Review 2 short movie clips while answering questions
Classroom Measurements Results Conducted successfully Feedbacks from students are positive
Classroom Measurements Observations 469,778 packets tx over the channel 99.3% of UDP and 0.7% of TCP 0.025% CRC errors 0.43% MAC retransmissions First 200 seconds: TCP (RTSP) UDP (RTP) data packets After 10 minutes: TCP (HTTP) Heaviest demand from multimedia streaming Max throughput is about 4.6 Mbps (vs. 5-6 Mbps effective BW of 11b) TCP (RTSP) UDP (RTP) TCP (HTTP) UDP (RTP) 4.6 Mbps Time
Additional Experiments 8 clients saturate the network; and the 9 th client degrades the performance of all clients equally. Bottleneck – the wireless channel
The Bad Apple Phenomenon Server Client 1Client 2Client 3
The Bad Apple Phenomenon Server Client 1Client 2Client 3 X
The Bad Apple Phenomenon Server Client 1Client 2Client 3 X
The Bad Apple Phenomenon Server Client 1Client 2Client 3 X One Bad Apple Spoils the Batch!
The Bad Apple Phenomenon
Server Wireless Network Interface Queue Wireless Channel MAC Protocol Client 1 Client 2 Client 3 Drop X
The Bad Apple Phenomenon
ScenarioNormal 4 Clients 1 Bad, Retry = 16 1 Bad, Retry = 1 Video rate (fps) Audio rate (fps) Avg skipped video frames (per client statistics) 0(0/0/0/0)0(0/0/0/0) 36 (37/34/27/46) 2(0/0/0/7)2(0/0/0/7) Avg dropped packets (per client statistics) 0 (0/0/0/2) 330 ( 309 / 320 / 294 / 399 ) 80 ( 80 / 77 / 83 / 729 ) Avg lateness (sec) Playback Stats at Clients
Conclusions Portable Wireless Classroom Network is a promising technology for education Currently, b network can support limited number of streaming clients, and the bottleneck is at the wireless channel When system is overloaded, degrading performance for all clients The “bad apple” phenomenon can seriously degrade multimedia delivery in wireless environments; can a security problem
Current and Future Work More elegant solutions to the “bad apple” problem Adaptive MDMI (Multiplicative Decrease and Multiplicative Increase) algorithm Multi-Channel Multi-Rate algorithm QoS support in Wireless LAN -- keep “bad apple” problem in mind
QUESTIONS?