Presentation on theme: "A Technical Analysis of the Adaptive Bit Rate Technology in DASH – Dynamic Adaptive Streaming over HTTP Brian Bresnahan,"— Presentation transcript:
A Technical Analysis of the Adaptive Bit Rate Technology in DASH – Dynamic Adaptive Streaming over HTTP Brian Bresnahan, firstname.lastname@example.org@fas.harvard.edu Harvard University Extension School CSCI-E139 Spring 2013, Len Evenchik
Basic Network Diagram and Data Flow
Fundamental Problem How can you maintain seemless video and audio playback over a network that can not guarantee quality of service ? Solution: A streaming algorithm that adjusts to the bandwidth available on the network.
Video Streaming Technologies Adobe Flash progressive download Apple HTTP Live Streaming Microsoft Smooth Streaming (Netflix) Adobe HTTP Dynamic Streaming NEW! DASH – Dynamic Adaptive Streaming over HTTP
Empirical Network Analysis of Video Streaming Providers Video Company Proto- col CDNPlayer TechnologyNotes HuluHTTPAkamaiAdobe Flash YoutubeHTTPnoAdobe FlashHTTP GET of a file of mime-type video/x-flv. For 360p, every 6-10 seconds a 1.7 MB file is downloaded. At 720p, player continuously downloads using HTTP a large file with mime-type video/mp4. e.g. 53 minute video is 1.3 GB NetflixHTTPLimelight or Netflix Microsoft Smooth Streaming and Silverlight Silverlight is the player software technology. Packet capture analysis done with Wireshark and Google Chrome browser Developer tools on my home network.
Video Codecs and Bitrates CodecBitrate Resolution Width (pixels) Resolution Height (pixels) Notes AVC1256 Kbps320240 AVC1512 Kbps320240 AVC11 Mbps640480 AVC11.384 Mbps640480 AVC11.5 Mbps1280720 AVC12.0 Mbps1280720 MPEG-27.0 Mbps720480 DVD resolution. (playback bitrate from PS3) AVC122 Mbps19201080 Blu-Ray DVD resolution. (Playback bitrate from PS3. May burst up to 40 Mbps) ??38402160“2K” UHDTV. Also “4K”.
Adaptive Bitrate Algorithm A video streaming system should dynamically adapt to the network conditions and playback device CPU load then automatically select the highest possible resolution the network can support. The general concept is called “adaptive bitrate streaming” (ABR). It was conceived by the DVD Forum in 2002 (Wikipedia, Adaptive Bit Rate). ABR follows this simple algorithm: 1.Encode the media at various resolutions, color depth, frame rate. The result is multiple data sets with different sizes which in turn will require different network bandwidths for streaming for playback. 2.As the media is played, the client monitors the playback. If the playback stalls, use a lower bitrate encoding. 3.If the playback is not stalling and there is bandwidth available such that the player can use a higher bitrate encoding, switch to the higher bitrate encoding
DASH Specification The specification was promoted by the Motion Pictures Expert Group (MPEG), “a working group of ISO/IEC with the mission to develop standards for coded representation of digital audio and video and related data.” (http://mpeg.chiariglione.org) DASH under development since 2009. Dynamic Adaptive Streaming over HTTP (DASH) ISO Specification Number 23009-1
DASH Architecture Elements HTTP/Web Servers, not proprietary servers HTTP – Hypertext Transfer Protocol – Based on TCP, a reliable protocol that guarantees byte order – Already has encryption standards – Is stateless, making the server simple. Forces the complexity to the client where ABR algorithm needs to run. – Allows for “partial GET” XML based Manifest.MPD file Goal: support Live Event streaming as well as static encoded media
DASH Network Model
Operational Steps 1.A back-end process is used to prepare and encode the media. This is the box on the left. This needs to be done before any clients can view the media. 2.The DASH client requests from the DASH server a “manifest” XML file called the MPD (Media Presentation Description) file. This file contains information regarding all available bitrates i.e. “Representations” for a particular video. 3.The media playing software application uses the DASH client to select a video and audio bitrate from the manifest and begins downloading both separately with HTTP. 4.The DASH client is codec agnostic so it passes the received video frames and audio packets to the media application to render them with the codecs on the operating system the application is running on. 5.The DASH client runs the ABR algorithm previously described. If the video or audio segments are downloading well, the next higher bit rate may be switched to. Similarly, if the segment download rate is insufficient for the chosen bitrate, a lower bitrate stream will be selected. 6.The DASH client repeatedly does HTTP GETs for segments of video and audio data until the video and audio playback is complete.
DASH Data Model Elements
Media Stream Access Points Required for the DASH client to be able to seamlessly switch bitrates as network bandwidth increases or decreases.
Media Presentation Description File http://cdn1.example.com/ 7657412348.mp4 3463646346.mp4 streamer.harvard.edu/e139_vid_2_1mbps.mp4 streamer.harvard.edu/e139_vid_2_2mbps.mp4
Use of URLs DASH supports three main methods of retrieving video and audio data using URLs. By specifying an exact URL to a file. It needs to be retrieved with an HTTP GET. In this approach, the DASH client must get the entire file. The files are explicitly named in the MPD file. By specifying an exact URL to a file with expectation that an HTTP partial GET will be used to retrieve data from a start byte to an end byte. In this approach, the DASH client can repeatedly read different parts of the same file. By a method called “Template based Segment URL Construction” (DASH Spec p. 48). In this method, variables $RepresentationId$ and $Number$ are placed in the MPD file. The DASH client uses them to dynamically create the URL needed to make a request for.
URL Method 2 Example This example is of method 2 described above. The MPD has 2 separate URLs for retrieving 1 Mbps encoded or 2 Mbps encoded media from the ficticious server streamer.harvard.edu: streamer.harvard.edu/e139_vid_2_1mbps.mp4 streamer.harvard.edu/e139_vid_2_2mbps.mp4 Note that since the Representation does not describe a variable name for the URL, then the expectation is HTTP partial GETs using begin and end bytes will be used to retrieve segment data.
URL Method 3 Example CSCIE_139_Lecture5_ From the above BaseURL tag and Segment Template, the DASH client can synthesize the following URLs over time: CSCIE_139_Lecture5_1130kbps_00000.ts CSCIE_139_Lecture5_1130kbps_00001.ts …
Other DASH Features Metric Reporting – the DASH spec defines 3 “observation points” that can be used to monitor the DASH client retrieving segments with HTTP and feeding the data to the client applicaton. (ISO, DASH Specification, p. 107) Switchable and selectable streams – useful to select different language audio streams or different camera angles if available. Ad insertion – DASH defines the ability to insert ads between Periods or Segments. Multiple Base URLS – the Representation can specify multiple URLs to allow the client to use different servers or CDNs if needed Clock drift control – for live sessions, the UTC time can be included with each segment to allow the client to adjust playback to match the server’s encode rate.
Youtube/Google DASH Demo Page
Bitrate Transition in Youtube DASH Log 31: DASH MSE/EME demo version 0.2-165-gfee2015 31: -------- Initializing -------- 35: 1, XHR for manifest sent 230: 2, Manifest received 393: 3, onSourceOpen() 398: 4, updateRepresentationForm() (AUTHOR’S NOTE: the “87” file is initial video file and “8c” is the audio file.) 896: 2, Sent XHR: url=http://yt-dash-mse-test.commondatastorage.googleapis.com/car-20120827-87.mp4, range=bytes=0-1183 899: 2, Sent XHR: url=http://yt-dash-mse-test.commondatastorage.googleapis.com/car-20120827-8c.mp4, range=bytes=0-851 (AUTHOR’S NOTE: REMOVED PART OF LOG. Note the next line shows the player selecting a video file with a different bit rate.) 1899: Selected new rate 265000 for bandwidth 948028.9207168575 (from 948028.9207168575, 1387512.245929166) 1900: 3, onRepChange 1904: 1, resetSourceBuffer 2396: 2, Sent XHR: url=http://yt-dash-mse-test.commondatastorage.googleapis.com/car-20120827-85.mp4, range=bytes=0-1147 2397: 2, Sent XHR: url=http://yt-dash-mse-test.commondatastorage.googleapis.com/car-20120827-8c.mp4, range=bytes=159832-318918
Conclusion DASH exemplifies the engineering of the internet. That is, it is built on previously designed and proven building blocks. DASH is built largely on HTTP and also on XML. Beneath HTTP is the TCP protocol delivering an error- free, ordered byte stream required for quality video display and audio playback. HTTP and TCP are available in a wide variety of devices today from handheld computers such as iPhones to HD TVs, creating the possibility of DASH being deployed on this wide variety of devices. The design of DASH appears to be solid.