<<

Sky Italia - Operation Evolution London – March 20th, 2018 1 Sky to IP-based distribution

Content Contribution Core Network Access Network Home Content Display Transmission Network (FTTx) Network

Public Internet / Global CDNs

Managed CDN Headend

Headend Managed CDN

Private CDN

Headend

2 Manage the entropy of IP CDN Selector

BI Daily Realtime Logs Macro Alerts Conviva CRM Tech Reporting

Off-line Telemetry Real–time Tracking Beacon CDN Logs Intelligence Logs

1 minutes hearbeat

Distributed over 2 datacenters and 2 public clouds 4 catalaogues made Original Personalized with 3 streaming protocols and 3 Manifest Manipulated Manifest DRMS CDN SELECTOR

Encoding superset of bitrates for linear, VOD and Origin Client D&P Content Content Server devices Chunks Chunks CDNs

7 CDNs: ISP local, global and private

4 Define a QoS vQoS and Sky Service Priority

• Provide high quality experience during multiple event of live-vod streaming + 1 or more concurrent streaming events. Lower priority services are limited via:

• Simulation of network QoS experience • Throttling of stream delivery on edge server • Midstream switching between same service/different capped bandwidth FQDN, based on telemetry information • Collection of real time concurrent activity via telemetria

• Priority list: VOD/Linear streaming, OTT streaming, Recording/D&P

6 vQoS and Sky Service Priority

7 Improve the CDN distribution capabilities New CDN Ranking and Allocation Logic High-Level Functioning of the Algorithm

The current prototype algorithm routes a given IP aggregation at given time following 3 main steps. Input variables to the algorithm can be weighed as wished.

1 2 3

Quality of Service QoS-driven QoS-driven Index CDN ranking allocation Bitrate Number of errors Rebuffering events & time

The output of the algorithm for a given IP aggregation (in this case IP /20) and time instance is the following table reporting the fundamental results of the calculations. CDN # SSTT Events QoS Index Final Allocation CDN 1 22 0.6630 25.64% CDN 2 18 0.5549 21.46% CDN 3 46 0.5345 20.67% CDN 4 1 0.4000 15.47% CDN 5 73 0.2187 8.46% CDN 6 0 NA 8.30% 10 Cloud Scalability Video Encoding Workflow as it was

‘Standard ’ VOD (on STB, Box sets, Kid Apps, , Nowtv)

on- Transcoding Multiplexing Packaging Encryption Distribution CDNs premise VOD solution

Content • Persistent File • Persistent Clear • Temporary Clear • OTT Catalogue • 8-15Mb/s Cache Packaged Cache • HLS, HSS, Dash, Ingest • 1080i H264 • Multi bitrate mp4 • HLS, HSS, Dash, TS Content Master File • mp4 @8-15 Mb/s • 1080p up to 5Mb/s TS • 1080p up to 5Mb/s • 50Mb/s • 1080p up to 5Mb/s • High Resolution/ MPEG2

12 Video Encoding Workflow Evolution

‘Standard’

cloud Cloud platforms • Sky Technology DC Transcodin VOD • Amazon Web Service g solution • Google Cloud Engine

‘Standard’ VOD (on STB, Box sets, Kid Apps, Sky Go, on- NowTV) Transcodin Multiplexin Packaging Encryption Distribution CDNs premise g g VOD

solutionContent Ingest • Persistent Clear Cache • Temporary Clear • OTT Catalogue Content Master File • Multi bitrate mp4 Packaged Cache • HLS, HSS, Dash, TS • 50Mb/s • 1080p up to 5Mb/s • HLS, HSS, Dash, TS• 1080p up to 5Mb/s • High Resolution/ MPEG2 • 1080p up to 5Mb/s

13 Video Encoding Workflow: Sky Italia Solution

Sky Italia VOD Hybrid solution can reduce the 2nd Step Cloud platforms lead time for Transcodin • Red Hat Openshift delivering VOD g • Public Cloud Providers assets while

keeping high VOD (on STB, Box sets, Kid Apps, Sky Go, Nowtv) signal quality 1st Step Transcodin Multiplexing Packaging Encryption Distribution without g increasing costs

Content Ingest • Persistent File • Persistent Clear • Temporary Clear • OTT Catalogue Content Master File • 8-15Mb/s Cache Packaged Cache • HLS, HSS, Dash, • 50Mb/s • 1080i H264 • Multi bitrate mp4 • HLS, HSS, Dash, TS • High Resolution/ • mp4 @8-15 • 1080p up to 5Mb/s TS • 1080p up to MPEG2 Mb/s • 1080p up to 5Mb/s 5Mb/s

14 Low Latency Low Latency

• A self developed origin server threats the streaming signal to be served with a minimal delay to live • It works with any standard players • Processing: http://lowdelay.sicdn.it/getFile/ • 2 chunks look-ahead manifest 217851/tg24/index.m3u8?align • 2s chunks =always • Serve-before-close file delivery • All clients aligned to the same chunk • More aggressive processing and latency could be reached with dedicated players

16 Video Artificial Intelligent Binge Watching

To detect the opening and closing credits in a video and also discriminates non interesting Next Episode in 5 S1: E 12 “Two credits from storytelling credits Swords” which are slightly different from PLAY episode to episode and are considered part of the show by viewers.

Episode 11 Episode 12

18 Thanks to video analytics is possible to extract more info… Video Content Recognition

FACE NUDE METADATA VIDEO RECOGNITION DETECTION EXTRACTION ANALYSIS Applying objects Artificial Intelligence recognition Algorithms may be technology can be used functionalities to videos, AI technology can be defined and used to train to add celebrity it’s possible to detect applied to analyze Sky AI systems starting from tagging and face and tag particular types videoclips (or coming human / subjective detection capabilities to of scenes based on from 3rd parties), to evaluations and existing systems, in specific objects they automatically detect categorizations, aiming order to let the may contain (e.g., those with inappropriate to assess “how much” a automatic recognition of «violent» sequences content containing nude video can be classified celebrities during the may be detected in a scenes, to be discarded as “violent”, or “comic”, playback of usual video movie, when «shoots» / censored. or belonging to other contents of the or «blood» are categories. distribution pipeline. recognized in those scenes) VIDEO CONTENT RECOGNITION APPLICATIONS

19 20 Quality Optimization

Optimization @Constant Optimization @Constant Quality Bitrate Constan Constan • Objective: optimizing the t t • Objective: maximizing the Bitrate Quality quality/bitrate ratio. Providing a perceptual quality, Approac Approac controlled quality in content- maintaining the classical h h aware VBR to reduce costs and constraints of CBR encoding optimize QoE

21 22