Labmeeting - 20150831 - Overhead and Performance of Low Latency Live Streaming using MPEG-DASH
Download as PPTX, PDF1 like724 views
This document summarizes research into reducing latency for live video streaming using MPEG-DASH. It introduces MPEG-DASH and how using HTTP chunked transfer encoding and Gradual Decoding Refresh encoding can help lower latency compared to basic DASH. The paper describes experiments conducted to generate and distribute live content using these techniques and evaluate latency, finding they were able to achieve latency as low as 240ms.
Labmeeting - 20150831 - Overhead and Performance of Low Latency Live Streaming using MPEG-DASH
- 1. NTUST - Mobilizing Information Technology Lab Overhead and Performance of Low Latency Live Streaming using MPEG-DASH Bouzakaria, N.; Concolato, C.; Le Feuvre, J. The 5th International Conference on Information, Intelligence, Systems and Applications, IISA 2014, Advisor：Jenq-Shiou Leu Student：Bing-Syuan Wang Date：2015/8/31 National Taiwan University of Science and Technology
- 2. NTUST - Mobilizing Information Technology Lab Outline • Introduction • Related Works • Live Dash Streaming Latency • Experimentations and Results • Conclusion 2
- 3. NTUST - Mobilizing Information Technology Lab Introduction • HTTP Streaming technologies have been introduced recently to deliver multimedia streams. • An important standard in the field of HTTP streaming is the MPEG Dynamic Adaptive Streaming over HTTP (DASH) standard. • HTTP/TCP vs RTP/UDP 3
- 4. NTUST - Mobilizing Information Technology Lab Introduction • DASH is not initially adapted for low latency. • Using “chunked-transfer encoding” over HTTP 1.1. • Using low latency video coding tools. -H.264/MPEG-4 AVC standard. 4
- 5. NTUST - Mobilizing Information Technology Lab Related Works • An HTTP streaming chain with typical segment duration of 1s would introduce a delay in the order of 3s compared to RTP. • With chunked-transfer encoding and with a proper download strategy, the latency depends on the duration of the HTTP chunks. • The average bitrate loss of GDR(Gradual Decoding Refresh) compared to periodic IDR(Instantaneous Decoding Refresh) was between 11 and 17% 5
- 6. NTUST - Mobilizing Information Technology Lab Related Works • Using the ISO Base Media File Format (lSOBMFF) in HTTP streaming systems, including DASH. • The goal of this paper: HTTP streaming using DASH with the combined usage of HTTP "chunked- transfer" encoding, GDR encoding and the ISO Base Media File Format. 6
- 7. NTUST - Mobilizing Information Technology Lab Live DASH Streaming Latency • Basic DASH Latency -Client, MPD(Media Presentation Description), MPD@availabilityStartTime • Low Latency DASH -Segments are further divided in smaller parts and these parts delivered using HTTP chunks -availabilityStartTimeOffset (ASTO) 7
- 8. NTUST - Mobilizing Information Technology Lab Live DASH Streaming Latency • Segment duration (ds) • Fragment duration (dc) • λ represents a margin introduced in the computation of ASTO to cope with UTC mismatch between the client and the server ASTO = AST − (AST − ds + dc + λ)=ds − dc − λ 8
- 9. NTUST - Mobilizing Information Technology Lab Experimentations • 1) Content generation: -encoding, GDR, ISOBMFF, MPD • 2) Content distribution: -”chunked-transfer” encoding • 3) DASH client: -processing incomplete segments 9
- 10. NTUST - Mobilizing Information Technology Lab Results 10
- 11. NTUST - Mobilizing Information Technology Lab Results 11
- 12. NTUST - Mobilizing Information Technology Lab Conclusion • HTTP Streaming is the new approach for streaming video over the Internet, for live and on demand cases. • This paper described an implementation of a streaming system comprising a DASH live encoder generator, a DASH-aware web server and a DASH client. • This paper validated the approach for very low latency streaming in local networks, with latency as low as 240 ms. 12
Editor's Notes
- #4: 雖然基於數據包的流媒體解決方案，例如使用RTP，可以實現延遲，在幀的順序，沒有今天用於這種非常低的延遲流HTTP流傳輸解決方案，例如DASH。
- #5: video conferencing or live streaming with voting. 通常，HTTP應答消息中發送的數據是整個發送的，Content-Length消息頭欄位表示數據的長度。數據的長度很重要，因為客戶端需要知道哪裡是應答消息的結束，以及後續應答消息的開始。然而，使用分塊傳輸編碼，數據分解成一系列數據塊，並以一個或多個塊發送，這樣伺服器可以發送數據而不需要預先知道發送內容的總大小。通常數據塊的大小是一致的，但也不總是這種情況。
- #6: 一些研究論文已發表有關現場或低延遲過串流HTTP。 segmentation process 2. fully produced 3. the download time of the segment does not describe what happens for shorter chunks. 本文然而，不考慮它的HTTP流媒體使用，不評估開銷高清晰度的序列。
- #7: report that the approach is inefficient for bitrates lower than IMbps.
- #8: 被延遲：the segmentation delay; the asynchronous fetch of media segments; the time to download the segments; and the buffering at the client side.
- #9: 被延遲：the segmentation delay; the asynchronous fetch of media segments; the time to download the segments; and the buffering at the client side.
- #10: encoding the video in realtime, in particular using the GDR scheme; segmenting and formatting the video segments according to the ISOBMFF; and generating the DASH Media Presentation Description (MPD). 範圍為1片段攜帶一個視頻幀到1片段攜帶整個段。 當一個新的片段被發布，該服務器發送該片段作為一個chunk。採用這種方法，該段的下載可以開始該段是完全準備好和發布前。