NEXT GENERATION IP VIDEO ARCHITECTURE TRENDS, CHALLENGES, & SOLUTIONS Weidong Mao Senior Fellow, Comcast June 20, 2019
THE RISE OF OTT IP VIDEO SUBSCRIPTION VOD SERVICE • Deep content library • Navigation, discovery, recommendation VIRTUAL PAY TV SERVICE • Live linear TV programming • Video On Demand and Cloud DVR ONLINE & SOCIAL MEDIA VIDEO • Short-form and long-tail content 2
TRADITIONAL VIDEO PROVIDERS GO ALL IP LIVE LINEAR CHANNELS • Broadcast networks, cable networks, PPV, PEG VIDEO ON DEMAND • Subscription VOD, Transactional VOD, and EST CLOUD DVR • Cloud recording/playback and download to go IN-HOME AND OUT-OF-HOME STREAMING DIGITAL AD INSERTION (DAI) EMERGENCY ALERT SYSTEM, CLOSED CAPTIONING, BLACKOUT 3
CLOUD DVR EXPANDS IN SCALE CLOUD DVR (DIGITAL VIDEO RECORDER) • Example deployment with 500GB per subscriber • ~200 hour of HD content at 5.5Mbps AVC ABR • Up to 6 simultaneous recordings per subscriber • In home streaming to multiple devices • Download to go • Trending to TB scale of cloud DVR storage per customer • Unique copy requirement drives storage I/O and capacity 4
ENABLING TECHNOLOGIES FOR IP VIDEO ENCODING (AVC, HEVC, AV1, VVC) STREAMING (HLS, DASH) STORAGE (ON PREMISE, CLOUD) TRANSPORT (HTTP, RELIABLE UDP) CONTENT DELIVERY NETWORK (HTTP PROXIES) DIGITAL RIGHTS MANAGEMENT (MULTI-DRM / MPEG CENC) DYNAMIC AD INSERTION (DAI) (SERVER & CLIENT SIDE) 5
IP VIDEO REFERENCE ARCHITECTURE 6
IP VIDEO – SMALL SCREEN IP VIDEO REACHES SECONDARY SCREEN OF ALL TYPES • U.S. adults spend more time on their mobile devices than in front of TV • OTT and managed video consumption increase significantly on mobile devices • Most devices support high end video format (e.g. 1080p60, 4K, HDR) 7
IP VIDEO – LARGE SCREEN YET PRIMARY SCREEN IS DOMINATING AND IMPROVING • Higher resolution: from HD 1920x1080 to 4K 3840x2160p and 8K 7680x4320 • Higher frame rate: from HD i60/p24/p30 to HD p60 to UltraHD p60/p120 • High Dynamic Range (HDR): more bits for brightness/contrast nonlinearly • Wide Color Gamut: more colors for richer video viewing experience 8
US TV HOUSEHOLD PROJECTION 4K and 4K HDR estimated projections (M) 140 120 100 80 60 40 20 0 2014 2015 2016 2017 2018 2019 2020 2021 2022 2023 2024 2025 2026 2027 Total US TVHH Total HD TVHH Total 4K TVHHs Total 4K HDR TVHHs Total Cable, DBS, Telco HD Subscribers Source: SNL Kagan, S&P Global Market Intelligence “US TV set and Blu-ray player projections through 2027” 87% of 2017 model year 4k TV’s have HDR support. 2018 model year and beyond 4k TVs all have HDR support 9
COMPRESSION (10X EFFICIENCY GAIN) HD 1080p60 1 0
MORE & BETTER TOOLS FOR VIDEO CODEC Codec MPEG-2 MPEG-4 AVC/H.264 HEVC/H.265 Partition size 16x16 Macroblock 16x16 Macroblock Coding Unit (64x64 to 8x8) Prediction unit Inter 16x16, 16x8, 8x16 Sub-block down to 4x4 Prediction Unit Quad Tree down Intra 8x8 to 4x4 Intra prediction DC predictor Up to 9 predictors Up to 35 predictors Inter prediction Motion vector from one Weighted prediction from Advanced Motion Vector reference picture multiple reference frames Prediction (spatial + temporal) Motion precision 1/2 pixel 1/2, and 1/4 pixel 1/4 and 1/8 pixel Transform Floating point DCT Integer DCT (8x8, 4x4) Integer DCT (32x32 to 4x4) Quantization Fixed weighted matrix Extended and finer control Extended and finer control Entropy coding Variable Length Coding Context Adaptive Binary Context Adaptive Binary (VLC) Arithmetic Coding (CABAC) Arithmetic Coding (CABAC) In-loop Filters None Deblocking Deblocking and SAO 1 1
CONTENT-ADAPTIVE ENCODING CONSTANT QUALITY, VARIABLE BITRATE • IP video delivery enables Variable Bit Rate (VBR) or Capped Variable Bit Rate (cVBR) • Aggressive VBR helps balance the need for quality, reach, and infrastructure capacity • Per-shot content-adaptive encoding shown to perform better than fixed bitrate ladders • More sophisticated player-side rate adaptation logic is needed Quality Video Segment #1 talking head Consistent Quality Video Segment #2 action Bitrate Equal Bitrate Allocation 1 2
VIDEO QUALITY METRICS PERCEPTUAL QUALITY METRICS ESSENTIAL FOR LARGE SCALE VIDEO SERVICES • Non-reference metrics as “gating criteria” for sources • Full-reference metrics to monitor and tune encoder performance • Full-reference metric as “target quality” for constant-quality encoding and streaming 1 3
ABR TREND – LOW LATENCY Source: https://www.wowza.com/blog/10-streaming-trends-2019 1 4
ABR TREND – FOCUS AREAS • Choosing the bitrate/resolution pairs to make up/downshifts least visible Content Preparation • Picking the best segment durations • Modeling streaming dynamics for different genres • Providing network conditions to streaming clients Distribution and Delivery • Achieving controlled fairness in the network • New transport options such as HTTP/2 • Modeling the impact of faster zapping and trick modes on the QoE QoE Modeling and Client • Understanding the impact of QoE on viewer engagement Design • Optimization across clients based on the stream utility • Understanding the interaction of adaptive streaming with caching in CDNs Analytics, Fault Isolation • Extracting actions based on real-time analytics and Diagnostics • Fixing issues faster and remotely 1 5
NETWORK IMPACTS OF IP VIDEO CONCURRENCY • Number of simultaneous viewers for popular live/on-demand content • High concurrency increases access network bandwidth consumption • High concurrency increases CDN caching efficiency, but also no. of connections RELIABILITY • Packet loss in various network links (backbone, regional, access, in-home Wifi) LATENCY • Live point latency, start up latency 1 6
VIDEO TRANSPORT OPTIONS MPEG-2 TS OVER UDP • Unreliable (UDP w/o FEC) • Extremely low latency • Multicast and unicast Reliability ABR OVER HTTP • Reliable (TCP) • High latency Latency • Unicast only CMAF ABR OVER CHUNKED HTTP Bandwidth • Reliable (TCP) efficiency • Low latency • Unicast NEXT GENERATION TRANSPORT? 1 7
EVOLUTION OF VIDEO TRANSPORT MPEG-2 TS over UDP Next Generation Video Transport Reliable High latency Unicast only ABR over HTTP (DASH, CMAF) 1 8
CHOOSE THE RIGHT STORAGE SOLUTIONS On-prem SSD On-prem HDD Cloud + SLA Cloud Storage HIGH THROUGHPUT MEDIUM PERFORMANCE HIGH AVAILABILITY & LOW LATENCY HIGH VOLUME ELASTICITY HIGH COST LOW COST HIGH LATENCY LOWEST COST Encoding / Packaging VOD library CDN Cache VOD Library Cloud DVR (cold) Cloud DVR (hot) Cloud DVR (warm) Metadata ~5GB/sec transfer speed ~250 MB/sec transfer speed 0.5-200 MB/sec transfer speed ~550K IOPs (r/w mix) ~80 IOPs SLA pay-per-performance ~20 μ s latency ~10 ms latency (incl. seek) 1 9
IP VIDEO STANDARDS Encoding Delivery Creation HTTP 1.1/2/3, TCP, UDP, NTP, RTP, IPV4, IPV6 MPEG AVC/H.264, MPEG HEVC/H.265, VVC, MPEG-H, MPEG DASH, SAND, MPEG CENC (Cbcs, Ctr), DASH-IF IOP, CPIX, DASH Events, ATSC 3.0 Indexing ISO-BMFF, CMAF, MPEG-TS, ST 2022, ST 2110, SCTE 128-1/2, 215-1/2/1-1, 214-1/2/3/4, 223 SCTE 35, 224, 250, 130 ATSC A300, A341, A342, A343, A334, A330, ETSI 101 154, DVB-DASH, IPTV BT.2100-2, ST 2086, PQ, HLG, IMF, IMSC, SMPTE 2094- 10, 2094-40, SL-HDR1, ST 2113, ISO/IEC 23091 CSTF, DPCTF, HATF, Streaming Media QoE, 864.G, HDMI. QUIC, XML, HTML 5, H.770, H.730, H.740, H.750, H.760, H.7770, H.780, TR 26.969, TR 26.939, TR 26.907, …. Playout 2 0
ML/AI APPLICATIONS TO IP VIDEO VIDEO COMPRESSION • Scene detection and classification • Content adaptive encoding CONTENT DELIVERY NETWORK • Intelligent caching algorithm ABR STREAMING • Improve ABR quality of experience CONTENT DISCOVERY • Voice remote based discovery • Smart video recognition and search BIG DATA ANALYTICS 2 1
DEPLOYING LARGE SCALE IP VIDEO Scalability Reliability Simplicity Multi-tenancy Multi-tenancy Multi-tenancy Security Velocity Cost Effectiveness Multi-tenancy Quality Operability Ubiquity 2 2
SUMMARY IP VIDEO IS BEING WIDELY DEPLOYED SCALABILITY, RESILIENCY, AND QUALITY ARE STILL A CHALLENGE ENABLING TECHNOLOGIES CONTINUE TO EVOLVE • Compression algorithms • Storage technologies • Network infrastructure • Transport protocols • Applications of ML/AI STANDARDS, OPEN SOURCE, INDUSTRY/ACADEMIA COLLABORATION 2 3
Recommend
More recommend
