Application, Network, and Link Layer
Measurements of Streaming Video
over a Wireless Campus Network
Group Members:
Berk BerkerEmrah BayraktarogluS. Tuncer ErdoganMustafa Omer KilavuzErkan Okuyan
I. Introduction (The Reason)The Reason for the Expectations about
Streaming Video:• Decrease in price of W-LAN AP’s• Increase in wireless link capabilities up to
54 Mbps
I. Introduction (The Problem)The problem is all about:• RealNetworks & Windows Streaming Media
make decisions, but it is unclear that:– Frame Lost Rate?– Signal Strength?– Link Layer Bitrate?
are more important?
I. Introduction (4 Hypothesis)This measurements study considers:i. W-LANs make it difficult for streaming
video to gracefully adapt when network conditions degrade.
ii. Multiple level encoding can stream better than videos encoded with only a single level when W-LAN conditions are poor.
iii. TCP is more effective than UDP.
iv. Current available estimation techniques for capacity are inadequate for W-LAN
II. Methodology (Tools)1. Application Layer:• Media Tracker collects application layer
data specific to streaming video including:– Encoding data rate– Playout bitrate– Time spend buffering– Video frame rate– Video frames lost– Video frames skipped– Packets lost– Packets recovered
II. Methodology (Tools)2. Network Layer:• UDP Ping measures:– Round-Trip time (?)– Packet loss rate along the stream flow
path
by providing:
– Constant ping rates– Configurable ping intervals in
milliseconds– Configurable ping packet sizes
II. Methodology (Tools)3. Wireless Data Link Layer:• WRAPI library was enhanced to collect
information about:– Signal strength– Frame retransmission counts and
failures– The specific W-AP that handles the
wireless last hop to the client
• Typeperf collects:– Processor utilization– Various network data
II. Methodology (Experiment)Experiments are done with hardware:• Windows Media Server• Windows Media Service v9.0• Dell laptop (Centrino Mobile CPU, Windows
XP SP1, IEEE 802.11g Wireless Network Adapter)
• Airspace APs, providing IEEE 802.11a/b/g wireless service.
II. Methodology (Experiment)Experiments are done with videos of:• Two Video Clips: Coast Guard & Paris (Both
352x288 resolution & 30 frames per second, two minutes long)– Coast Guard: High Motion (5.4% skipped macro
blocks)– Paris: Low Motion (41.2% skipped macro blocks)
II. Methodology (Experiment)Experiments are done with:• Single-Level version of videos encoded at
2.5 Mbps to stress the wireless link• Multiple-Level version including 11
encoding layers• Streamed using TCP & UDP for comparison
II. Methodology (Experiment)Campus Network Map:
II. Methodology (Design)Experiments done:• Downloading a large file with wget
– To estimate the effective throughput of a TCP bulk transfer
• 2 clips x 2 versions x 2 transport protocols• A final bulk download• UDP pings to determine round-trip time and
package lost– 200 milliseconds apart– 1350-byte packets for single level video– 978-byte packets for multiple level video
II. Methodology (Design)Experiments done:• Five times x Three distinct locations x
Three floors in the CS department
= 45 experimental results
= 360 video streams
• Locations: Three laptop in good, fair, and bad reception locations
III. RESULTS• Collected data is:
• No significant statistical difference between the high-motion and the low-motion video.
• High-motion and low-motion does not have a significant effect on wireless network performance.
Categorization
• There is a “cliff” between signal strengths -70 and -80 dBm
Categorization
• From now on, experiments are categorized in one of the regions: “Good”, “Edge” or “Bad”
First Analysis: Single-Level vs. Multi-Level Encoding
• Multi-Level or Single-Level Encoded
• Streaming of single and multi-level encoded videos are compared according to their average frame rate in “Good” and “Bad” locations.
Single-Level vs. Multi-Level
Single-Level vs. Multi-Level
Single-Level vs. Multi-Level
• In a Good location, number of encoded levels have a very little effect, since the stream does not have to be scaled to a lower bitrate.
• In a Bad location, for the 2/3 of the time, multiple level stream has a higher frame rate than the single level one. (22 fps to 11 fps on average)
TCP Streaming Over UDP Streaming
• In good wireless locations, TCP streaming and UDP streaming have almost the same performance
• In bad wireless locations choosing one of them has significant impact on performance
TCP Streaming Over UDP Streaming
• In bad wireless locations choosing TCP Streaming(24fps) provides better frame rate than the UDP(15fps) streaming
• TCP Streaming also have lower coefficient of variation of frame rate than the UDP Streaming
TCP Streaming Over UDP Streaming
TCP Streaming Over UDP Streaming
TCP Streaming Over UDP Streaming
• TCP Streaming has better frame rates because TCP retransmits the data
• But, Without Built-in retransmissions,UDP does not recover the lost data, so loss rates occur
TCP Streaming Over UDP Streaming
TCP Streaming Over UDP Streaming
TCP Streaming Over UDP Streaming
• UDP uses a high data rate to fill the playout buffer
• AP queue grows long and AP cannot drain the queue, because wireless layer capacity is limited.
TCP Streaming Over UDP Streaming
TCP Streaming Over UDP Streaming
• TCP may have longer play out than the UDP for the same length of video
• Because in TCP, retransmissions take a lot of time
TCP Streaming Over UDP Streaming
The Challenges of Streaming over Wireless
TCP-Friendly Capacity
• s = packet size• R = round-trip time• p = packet drop rate
• trto = TCP retransmission timeout
Average Application Encoding Rate versus Wireless Capacity for TCP and UDP Streams
Average Application Encoding Rate versus TCP-Friendly Capacity for TCP and UDP Streams
Average Application Encoding Rate versus Wireless Capacity for Multiple and Single Level Stream
Average Application Encoding Rate versus TCP-Friendly Capacity for Multiple and Single Level Stream
Conclusion (4 Hypothesis) Revisited This measurements study considers:i. W-LANs make it difficult for streaming
video to gracefully adapt when network conditions degrade.
ii. Multiple level encoding can stream better than videos encoded with only a single level when W-LAN conditions are poor.
iii. TCP is more effective than UDP.
iv. Current available estimation techniques for capacity are inadequate for W-LAN
Improvement Areas1. Identifying and adapting to
challenging wireless transmission situations.
2. Understanding packet and frame burst loss behavior.
3. Effective media scaling
4. Real Media and Quick Time researches.
End of Presentation
Questions?