ITU-T G.1021

Abstract

Audio and video applications currently used in packet networks implement client buffers (e.g., de-jitter buffers and play-out buffers) at the receiver side. The behaviour of these buffers is essential to the end-to-end performance of these applications. This Recommendation defines models for these client buffers for audio/video applications currently used in actual networks, in order to estimate the client state based on the dynamic network behaviour. <br>In this Recommendation, different buffer models are derived depending on the different applications taken into account. However, these buffer models may be based on common intrinsic characteristics: i.e., buffer size, service rate, initial delay and re-buffering delay. Example buffer performance metrics include peak/average/instantaneous buffer occupation. <br>In this Recommendation, buffer models may accept as input, either live traffic arriving on measurement points, packet capture (pcap) files, or post-processed files. Buffer models are expected to provide client states (e.g., codec play-out begin, buffer overflow, buffer dry out, play-out resume) associated with relevant timing information. As part of future work, these buffer models could be used to derive new performance parameters and associated statistics in order to complement the IP-layer metrics defined in Recommendation Y.1540. <br>It is out of scope for this Recommendation to define specific de-jitter buffers for use in actual client media stream delivery. <br>In this Recommendation the following applications, rate and buffer characteristics, and associated protocol stacks are considered: <br>a) Audio applications: <br>i) Ideally constant fill, constant play-out rate and relatively low bandwidth. <br>b) Video "low bitrate mode" quarter common intermediate format-quarter video graphics array (QCIF-QVGA) resolutions, mostly for mobile TV and streaming with the sub application areas: <br>i) Linear mobile TV over real-time transport protocol (RTP) (including mobile TV over a 3G mobile network with multimedia broadcast/multicast services (MBMS) and with unicast transport over RTP/UDP/IP). <br>ii) Multimedia streaming (including 3GPP packet-switched streaming service (PSS) with transport over RTP/UDP/IP). <br>iii) [IETF RFC 3984] variable bit rate, up to 2 Mbps, with separate buffers for audio and video, and separate play-out rates for each buffer. <br>c) Video "high bitrate mode" standard-definition (SD) and high-definition (HD) television, mostly for Internet Protocol television (IPTV) with the following sub-application areas: <br>i) Linear broadcast TV (including transmission over MPEG2-TS/RTP/UDP/IP, MPEG2-TS/UDP/IP and RTP/UDP/IP transport). <br>ii) Video on-demand (including transmission over MPEG2-TS/RTP/UDP/IP, MPEG2-TS/UDP/IP and RTP/UDP/IP transport). <br>iii) [IETF RFC 2250], constant bit rate (CBR) or variable bit rate (VBR) up to 50 Mbps, with separate buffers for audio and video, and separate play-out rates for each buffer. <br>d) Over-the-top video applications: <br>i) Non-adaptive streaming (MPEG4_fileformat/HTTP/TCP, HTTP progressive download, HTTP partial download, chunks): <br>&#8211; Progressive download: TCP transport, TCP flow control (65Mbps super-peak rates, peak rate depends on network conditions and path), classic single buffer, leaky bucket at constant frame rate. <br>ii) Adaptive streaming dynamic adaptive streaming over HTTP (DASH/TCP): &#8211; Chunk Delivery: (e.g., DASH) bursts of packets, like sequential FTP file transfers.