MediaPlayer™ vs. RealPlayer™ A Comparison of Network Turbulence
Mingzhe Li, Mark Claypool, Robert Kinicki CS Department Worcester Polytechnic Institute Worcester, MA, USA http://perform.wpi.edu/ Introduction • Streaming video often does not like TCP – Wants smooth stream, so rate-based – Does not always want retransmissions !Chooses UDP • UDP flows may be unresponsive to congestion !Handle with Active Queue Management (AQM) • Typical AQMs model UDP flows as CBR (“firehose”) • More realistic models of streaming UDP flows will make AQMs more effective • We investigate size and distribution over time ! Turbulence • Big 3: RealPlayer, MediaPlayer, QuickTime [Jup01]
November 2002 SIGCOMM Internet Measurement Workshop Methodology
• Build automated video players (MediaTracker and RealTracker [WCZ01]) – Use commercial cores – Software Development Kits (SDKs) – Record application stats: frame rate… • Select hosts with both RealNetworks and Microsoft Media video servers • Stream identical content to players on the same host • Analyze results
November 2002 SIGCOMM Internet Measurement Workshop Type Bandwidth (Kbps)Clip (Kbps) Selection Real Microsoft 1 Sports High 284 323 3:46 Low 36 50 2 Commercial High 268 307 0:39 Low 84 102 3 Sports High 284 307 0:60 Low 37 38 4 Music T.V. High 181 309 4:05 Low 26 50 5 News High 218 250 1:47 Low 22 39 6 Movie Clip Very High 637 731 2:27 High 271 347 Low 39 102
November 2002 SIGCOMM Internet Measurement Workshop Experimental Setup
• Player Platform – Microsoft Windows 2000 PC – P4 1.8 GHz, 512M RAM, AGP 32MB video card – mindspeed.wpi.edu (Worcester, MA, USA) • Network – LAN: PCI 10M NIC Win2k professional – Campus uplink: ~50 Mbps (Yipes in 15 Mbps) • Software – Microsoft Media Player version 7.1 – RealNetworks RealOne Player build 6.0.10.505 –Etherealversion 0.8.20
November 2002 SIGCOMM Internet Measurement Workshop 1760 IP Packet Arrivals 830
1755 Real Play (217K) 825
1750 Windows Media Player (250K) 820
1745 815
1740 810
1735 805
Real Packet Number Packet Real 1730 800 Media Packet Number
1725 795
1720 790 30 30.2 30.4 30.6 30.8 31 Time (Seconds) • RealPlayer – single packets • Media Player – groups of packets (3 in this example) – 2 at about 1500 bytes (typical MTU) – 1 “left-over” less than MTU November 2002 SIGCOMM Internet Measurement Workshop Media Player IP Packet Fragmentation
100% 90% 80% 70% 60% 50% 40% 30% 20% 10% IP Packets w/out UDP Header UDP w/out Packets IP 0% 0 200 400 600 800
Encoded Bandwidth (Kbits/s)
November 2002 SIGCOMM Internet Measurement Workshop Normalized UDP Packet Sizes 0.8
0.7 Real Player 0.6 Windows Media 0.5
0.4
0.3
0.2 Probability Density
0.1
0 00.51 1.52
Normalized Packet size Media Player – More “Constant Packet Size” RealPlayer – Less “Constant Packet Size”
November 2002 SIGCOMM Internet Measurement Workshop UDP Packet Interarrival Times 1 0.9 0.8 0.7 0.6 0.5 0.4 0.3
Cumulative Density 0.2 Real Player 0.1 Window Media Player 0
0 0.5 1 1.5 2 2.5 3 Normalized Interarrival Time Interval
Media Player – More “Constant Packet Rate” RealPlayer – Less “Constant Packet Rate”
November 2002 SIGCOMM Internet Measurement Workshop Playout Rate 700 Real Player (284K) 600 Real Player (36K) 500 Windows Media Player (323K)
400 Windows Media Player (49K)
300
200
Bandwidth (Kbits/sec.) 100
0 0 50 100 150 200 250
Time (seconds)
RealPlayer buffers at a higher rate
November 2002 SIGCOMM Internet Measurement Workshop Frame Rate 35
30
25
20
15 Real Player Frame Rate (fps) Frame Rate
10 Media Player
5 0 200 400 600 800 Average Bandwidth (Kbits/sec.) RealPlayer significantly higher at low bandwidth
November 2002 SIGCOMM Internet Measurement Workshop Summary • Microsoft Media Player typical CBR – Constant packet sizes and interarrivals • RealNetworks RealPlayer less CBR – Range of packet sizes and interarrivals – Buffers at up to 3 times playout rate • Media Player has significant IP frag. – 3-5 IP frags / UDP for broadband+ bwidth • Results can be used for more realistic streaming simulations and emulations • MediaTracker and RealTracker online: http://perform.wpi.edu/real-tracer/
November 2002 SIGCOMM Internet Measurement Workshop Future Work • Build simulated streaming application –In NS • Server side control – Clip encoding, Content type – (We have server set-up, preliminary results) • Bandwidth under congestion –TCP-Friendly? – (We have results for RealPlayer) • More players and protocols –QuickTime (QuickTracker)
November 2002 SIGCOMM Internet Measurement Workshop MediaPlayer™ vs. RealPlayer™ A Comparison of Network Turbulence
Mingzhe Li, Mark Claypool, Robert Kinicki CS Department Worcester Polytechnic Institute Worcester, MA, USA http://perform.wpi.edu/