doc.: IEEE /0764r1 Submission Jan 2009 Alex Ashley, NDS LtdSlide 1 Using packet drop precedence for graceful degradation Date: Authors:

doc.: IEEE /0764r1 Submission Jan 2009 Alex Ashley, NDS LtdSlide 2 Abstract As one of its objectives, the aa PAR has: –Graceful degradation of audio video streams when there is insufficient channel capacity, by enabling packet discarding without any requirement for deep packet inspection There are many issues to consider to meet this goal. The aim of this presentation is to highlight some of these issues and to propose a simple end-to-end solution.

doc.: IEEE /0764r1 Submission Jan 2009 Alex Ashley, NDS LtdSlide 3 Assumption The assumption in this presentation is that packet discarding for graceful audio video degradation only makes sense for UDP (and UDP-Lite) based streaming TCP has its own congestion control algorithms. Dropping frames will not provide graceful degradation, just cause retransmissions and probably trigger congestion avoidance algorithms

doc.: IEEE /0764r1 Submission Jan 2009 Alex Ashley, NDS LtdSlide 4 An MPEG-2 transport packet is 188 bytes long and contains data for one component (e.g. video, audio, data) When sending audio video data that is inside MPEG-2 transport packets, typically 7 TS packets are placed inside one UDP packet Most UDP packets (typically >95%) contain a mix of components Video Frame MAC header IP header UDP header (RTP header) UDP (RTP) Packets

doc.: IEEE /0764r1 Submission Jan 2009 Alex Ashley, NDS LtdSlide 5 Components of a Transport Stream Tuning1.01% Timing1.02% Video87.06% Audio6.57% Subtitles0.06% Data4.30%

doc.: IEEE /0764r1 Submission Jan 2009 Alex Ashley, NDS LtdSlide 6 Components of a Transport Stream (2) The tuning information (1.01%) and data components (4.3%) are continuously repeated in the transport stream Loss of these packets reduces the speed of channel change or use of interactive applications, but has no impact on audio video experience Therefore, these transport packets make good candidates for discarding when there is congestion –But they don’t normally occur on UDP packet boundaries –Application layer needs to be aware of UDP packing to place these components in their own UDP packets.

doc.: IEEE /0764r1 Submission Jan 2009 Alex Ashley, NDS LtdSlide 7 Elementary Stream RTP Another method of audio video streaming is to place individual components in their own RTP stream –The MPEG-2 Transport Stream and PES headers are removed –Each component is sent to a different port –Timestamp field in RTP header is used to synchronize components Selective packet dropping in RTP ES is much easier, because only one component per UDP packet –But at the moment not widely deployed due to increased receiver complexity

doc.: IEEE /0764r1 Submission Jan 2009 Alex Ashley, NDS LtdSlide 8 Forward Error Correction (FEC) To cope with UDP packet loss, some systems add FEC packets –E.g. when streaming IPTV over xDSL –E.g. Adding 16 UDP packets of FEC to every 204 UDP packets of audio video data. Any 16 UDP packets can be lost without information loss Example FEC schemes: –RFC 2733 RTP Payload Format for Generic FEC –Reed-Solomon –RFC 5053 Raptor FEC –RFC 5170 LDPC Staircase and Triangle FEC –SMPTE Pro-MPEG Code of Practice #3 r2 –ETSI TS DVB-AL-FEC Packet discarding very easy as any packet can be a discard candidate –Just don’t throw away too many in one protection period!

doc.: IEEE /0764r1 Submission Jan 2009 Alex Ashley, NDS LtdSlide 9 Scalable Video Coding The video encoding is partitioned in a manner that has a baseline and one or more enhancement layers –E.g. an SD baseline plus enhancement to make HD However, SVC is not widely deployed –Encoder complexity –Decoder complexity –Not necessarily bandwidth efficient It’s often more efficient to send two separately encoded streams!

doc.: IEEE /0764r1 Submission Jan 2009 Alex Ashley, NDS LtdSlide 10 So how does this relate to ? There are many ways in which audio video data is streamed over IP networks Some methods lend themselves to selective packet discarding, some do not Probably the best pragmatic approach is some simple signalling per UDP packet and let the application decide when it can provide discard hints

doc.: IEEE /0764r1 Submission Jan 2009 Alex Ashley, NDS LtdSlide 11 DSCP At layer 3, per hop behaviour defined by DSCP (TOS) field Typical PHBs: –AF - Assured Forwarding based upon service level agreement –BE - Best Effort –EF - Expedited Forwarding for low-loss, low-latency traffic –CS – Class selector preserves backward compatibility with IP- precedence scheme AF defines 3 levels of drop precedence –But does EF make more sense for audio video? –Some standards bodies & industry forums have already pre-defined DSCP tags for AV in the home DVB-IPI uses AF (video= or ) DLNA uses CS (video=101000) In conclusion, layer 3 tag might not help us –A layer 2 based tagging solution is needed

doc.: IEEE /0764r1 Submission Jan 2009 Alex Ashley, NDS LtdSlide ad VLAN Tags Drop Eligible Indicator Priority & DE

doc.: IEEE /0764r1 Submission Jan 2009 Alex Ashley, NDS LtdSlide 13 PCP field in C-Tag Format Management decision on which row to choose. All Ports on a LAN should select the same row.

doc.: IEEE /0764r1 Submission Jan 2009 Alex Ashley, NDS LtdSlide 14 Stream Classification Service dot11ShortRetryLimit is expanded by 32 entries –Allows lower latency for AV streams without impact on legacy traffic –Allows reduced retry counts for packets marked discard eligible Request to AP to signal a stream that wants to make use of packet drop precedence –Uses the frame classifier TCLAS Element to specify frame marking –Provides suggested drop and non-drop retry limits –Allows flexibility in packet signalling –No need for new frame signalling –Using 802.1ad C-TAGs allows end-to-end use Marked packets are dropped when AP’s buffers are exhausted or the retry limit is reached

doc.: IEEE /0764r1 Submission Jan 2009 Alex Ashley, NDS LtdSlide 15 Example AP Source device tags UDP packets with priority and drop eligibility (DEI). IEEE used to connect source to the AP Request SCS Destination makes an SCS request to the AP to inform it that the stream has DEI and the method of tagging AV stream When active, the AP can use the DEI tagging to choose packets to discard if conditions deteriorate.

doc.: IEEE /0764r1 Submission Jan 2009 Alex Ashley, NDS LtdSlide 16 Deep Packet Inspection “Deep Packet Inspection is a term used to describe the capabilities of a firewall or an Intrusion Detection System (IDS) to look within the application payload of a packet or traffic stream and make decisions on the significance of that data based on the content of that data.” “Deep Packet Inspection (DPI) (also called complete packet inspection and Information eXtraction - IX -) is a form of computer network packet filtering that examines the data and/or header part of a packet as it passes an inspection point, searching for protocol non-compliance, viruses, spam, intrusions or predefined criteria to decide if the packet can pass or if it needs to be routed to a different destination, or for the purpose of collecting statistical information. This is in contrast to shallow packet inspection (usually called Stateful Packet Inspection) which just checks the header portion of a packet.”

doc.: IEEE /0764r1 Submission Jan 2009 Alex Ashley, NDS LtdSlide 17 Conclusions Even without scalable video coding, it is possible to arrange for some UDP packets to be more readily sacrificed than others Not all audio video streams will be amenable to graceful packet discarding A simple addition to allows suitable streams to make use of graceful degradation

doc.: IEEE /0764r1 Submission Jan 2009 Alex Ashley, NDS LtdSlide 18 References Slides 4 to 12 in document IEEE /0717r1 provide a good background on MPEG video encoding Deep packet inspection – –