(ARC303) Pure Play Video OTT: A Microservices Architecture
Report
Share
•14 likes•5,495 views
![HTTP Dynamic Streaming — HDS
> GET /live/channel1.abst HTTP/1.1
< Seg1-Frag55
< Seg1-Frag56
[…]
> GET /live/channel1.abst HTTP/1.1
< Seg1-Frag1
< Seg1-Frag2](https://image.slidesharecdn.com/arc303-151009185015-lva1-app6892/75/arc303-pure-play-video-ott-a-microservices-architecture-39-2048.jpg?cb=1666906310)
![CDN-aware Asset Info Service
> GET /asset/31337 HTTP/1.1
< assetUrl: "http://cdn-a.alexjs.im/vod/31337.m3u8"
< adProvider: "alexjsAds"
< countryCode: "[im, sg, id]"](https://image.slidesharecdn.com/arc303-151009185015-lva1-app6892/75/arc303-pure-play-video-ott-a-microservices-architecture-49-2048.jpg?cb=1666906310)
(ARC303) Pure Play Video OTT: A Microservices Architecture
•14 likes•5,495 views
Report
Share
Download to read offline
"An end-to-end, over-the-top (OTT) video system is built of many interdependent architectural tiers, ranging from content preparation, content delivery, and subscriber and entitlement management, to analytics and recommendations. This talk will provide a detailed exploration of how to architect a media platform that allows for growth, scalability, security, and business changes at each tier, based on real-world experiences delivering over 100 Gbps of concurrent video traffic with 24/7/365 linear TV requirements. Finally, learn how Verizon uses AWS, including Amazon Redshift and Amazon Elastic MapReduce, to power its recently launched mobile video application Go90. Using a mixture of AWS services and native applications, we address the following scaling challenges: Content ingest, preparation, and distribution Operation of a 24x7x365 Linear OTT Playout Platform Common pitfalls with transcode and content preperation Multi-DRM and packaging to allow cross platform playback Efficient delivery and multi-CDN methodology to allow for a perfect experience globally Kinesis as a dual purpose system for both analytics and concurrency access management Integration with Machine Learning for an adaptive recommendation system, with real time integration between content history and advertising data User, entitlement, and content management General best practices for ‘Cloud Architectures’ and their integration with Amazon Web Services; Infrastructure as Code, Disposable and immutable infrastructure, code deployment & release management, DevOps and Microservices Architectures This session is great for architects, engineers, and CTOs within media and entertainment or others simply interested in decoupled architectures."
More Related Content
What's hot
What's hot(20)
Similar to (ARC303) Pure Play Video OTT: A Microservices Architecture
![[AWS LA Media & Entertainment Event 2015]: Security of Digital Media Content ...](https://cdn.slidesharecdn.com/ss_thumbnails/awscontentsecurity-151113175039-lva1-app6892-thumbnail.jpg?width=384&fit=bounds)
Similar to (ARC303) Pure Play Video OTT: A Microservices Architecture(20)
![[AWS LA Media & Entertainment Event 2015]: Security of Digital Media Content ...](https://cdn.slidesharecdn.com/ss_thumbnails/awscontentsecurity-151113175039-lva1-app6892-thumbnail.jpg?width=768&fit=bounds)
More from Amazon Web Services
More from Amazon Web Services(20)
Recently uploaded
Recently uploaded(20)
(ARC303) Pure Play Video OTT: A Microservices Architecture
- 1. © 2015, Amazon Web Services, Inc. or its Affiliates. All rights reserved. Alex Smith, ASEAN Media Solutions Architect, AWS Vidhya Narayanan, Director, Verizon onCue October 2015 Pure Play Video, Over-The-Top A Microservices Architecture in the Cloud ARC303
- 2. The Problem
- 3. Your Consumers Have Changed More content choices More devices More delivery methods New experiences
- 4. Your Business Has Changed More data Faster go to market Leaner teams Innovation and failure
- 5. What to Expect from the Session Content is king Building from scratch Battle-hardened lessons – Verizon onCue
- 6. Clarifications OTT, OVP, AVOD, TVOD, SVOD It doesn’t matter Microservices Buzzword? Specific scope, interoperable services Allow rapid innovation
- 7. Over the Top Platform
- 11. Processing & Management Content Production Content Distribution Content Storage Content Consumption
- 12. Processing & Management Shared IT Services NetworkInfrastructure OperationsSecurity Content Production Content Distribution Content Storage Content Consumption
- 13. Processing & Management Content Production Content Distribution Content Storage Content Consumption
- 14. Your Existing Physical Requirements HD-SDI Satellite transmission Local SAN HW transcoder
- 15. Your OTT Approach FTP / Accelerator Large NAS DAM / Workflow System
- 16. Storage and Ingest — Serverless AWS Import/ Export S3 Ingest Content provider
- 17. Storage and Ingest — Serverless AWS Import/ Export S3 Ingest Lifecycle policies S3 bucket Amazon Glacier Content provider Lifecycle policies S3 IA
- 18. Storage and Ingest — Serverless AWS Import/ Export S3 Ingest S3 notification AWS Lambda Lifecycle policies S3 bucket Amazon Glacier Content provider Lifecycle policies S3 IA
- 19. Storage and Ingest — Serverless AWS Import/ Export S3 Ingest S3 notification AWS Lambda Metadata extract & persist Amazon DynamoDB Amazon CloudSearch S3 bucket Amazon Glacier Content provider Lifecycle policies Lifecycle policies S3 IA
- 20. Storage and Ingest — Serverless AWS Import/ Export S3 Ingest For source, renditions, metadata sidecar files S3 notification Amazon Elastic Transcoder proxy / thumbnail generation AWS Lambda Metadata extract & persist Amazon DynamoDB Amazon CloudSearch S3 bucket Amazon Glacier Content provider Lifecycle policies Lifecycle policies S3 IA
- 21. Storage and Ingest — Serverless AWS Import/ Export S3 Ingest For source, renditions, metadata sidecar files API Gateway CloudFront distribution S3 notification Amazon Elastic Transcoder proxy / thumbnail generation AWS Lambda Metadata extract & persist Amazon DynamoDB Amazon CloudSearch AWS Lambda S3 bucket Amazon Glacier Content provider Lifecycle policies Lifecycle policies S3 IA
- 22. Processing & Management Content Production Content Distribution Content Storage Content Consumption
- 23. Encoding, Packaging, Encrypting Challenges Multiple client devices Higher quality content Parallel, complex workflow Uneven load distribution
- 24. Content Processing Pipeline (Using Lambda) AWS Import/ Export S3 Ingest Ingest
- 25. Content Processing Pipeline (Using Lambda) AWS Import/ Export S3 Ingest S3 notification Ingest
- 26. Content Processing Pipeline (Using Lambda) AWS Import/ Export S3 Ingest S3 as backend storage for content files accessible to other processing tasks Amazon Elastic Transcoder S3 notification Trigger a Lambda function to start a transcoding job Ingest
- 27. Content Processing Pipeline (Using Lambda) AWS Import/ Export S3 Ingest S3 as backend storage for content files accessible to other processing tasks Amazon Elastic Transcoder S3 notification Trigger a Lambda function to start a transcoding job Ingest S3 notification
- 28. Content Processing Pipeline (Using Lambda) AWS Import/ Export S3 Ingest S3 as backend storage for content files accessible to other processing tasks Amazon Elastic Transcoder S3 notification Trigger a Lambda function to start a transcoding job Ingest S3 notification Lambda function to extract metadata Update CMS or metadata
- 29. Content Processing Pipeline (Using Lambda) AWS Import/ Export S3 Ingest S3 as backend storage for content files accessible to other processing tasks Amazon Elastic Transcoder S3 notification Trigger a Lambda function to restart a transcoding job Ingest S3 notification Lambda function to extract metadata Update CMS or metadata API Gateway
- 30. What About Packaging? All moving towards HTTP – RTMP is hard to scale Battle of the standards – HLS, HDS, SmoothStream, MPEG-DASH Alliance for Open Media
- 31. And Encryption / DRM? Still largely driven by studio requirements Just-in-time encryption (hard) Reusability across packaging methods (PlayReady across HLS & SmoothStream)
- 32. Pitfalls of Content Prep Betting big on closed standards Technologies in vogue Adoption is device-driven
- 33. Avoiding Pitfalls of Content Prep Keep your mezzanine / masters (S3 IA, Glacier) Mix and match your encoding Contradictory — JIT packaging and heavy caching Remember 80/20 (95/5) rule of content
- 34. Top Tip CPU-based encoding c4.8xlarge / m4.10xlarge P-State, C-State configuration Varies with particular encoding libraries – experiment and baseline (Thanks PocketMath!)
- 35. Processing & Management Content Production Content Distribution Content Storage Content Consumption
- 36. Live RTMP ingest to origin Repackaged and encrypted on the fly DRM less common WebRTC for 1-to-few, then RTMP/HTTP when traffic increases Content Delivery Architectures VOD / Catchup File-based – stored on S3/CDN Usually prepared (encrypted/packaged) before final storage DRM by default ISPs worried about unicast delivery
- 37. HTTP Live Streaming (HDS & HLS) HLS was pretty close to de facto Space becoming disrupted again Network considerations in the real world
- 38. HLS and HDS Uses “parent / child / chunk” model HLS: Playlist contains chunklists, which contain chunks HDS: Manifest contains bootstrap files, which contain fragments
- 39. HTTP Dynamic Streaming — HDS > GET /live/channel1.abst HTTP/1.1 < Seg1-Frag55 < Seg1-Frag56 […] > GET /live/channel1.abst HTTP/1.1 < Seg1-Frag1 < Seg1-Frag2
- 40. HTTP Dynamic Streaming — HDS Use fragment name alignment in the event of a republish Ensure that fragments are aligned across bitrates Players will error otherwise
- 41. Top Tip HDS bootstraps are binary To view, you must decode them first https://bitly.com/abstdecoder (Thanks SwiftServe!)
- 42. HTTP Live Streaming — HLS > GET /live/channel1.m3u8 HTTP/1.1 < HTTP/1.1 200 OK < Date: Tue, 15 Sep 2015 13:37:56 GMT < Server: Apache < Last-Modified: Mon, 03 Aug 2015 3:14:15 GMT < Accept-Ranges: bytes < Content-Length: 219
- 43. HTTP Live Streaming — HLS > GET /live/channel1.m3u8 HTTP/1.1 < HTTP/1.1 200 OK < Date: Tue, 15 Sep 2015 13:37:56 GMT < Server: Apache < Last-Modified: Mon, 03 Aug 2015 3:14:15 GMT < Accept-Ranges: bytes < Content-Length: 219 < Cache-Control: max-age=5
- 44. HTTP Live Streaming Add a no, or short age, cache header Use segment name randomisation Use Player metrics to detect problem ISPs / CDNs
- 45. Multi-CDN
- 46. CDN Selector Build vs. buy Excellent offerings on the market Metadata often tightly coupled with platform, so tricky to use a third party
- 47. CDN Selection Methods DNS-based Geo / latency / intelligent routing at DNS level Asset sharding 50% of assets on CDN A, 50% on CDN B CDN-aware asset info service
- 48. CDN-aware Asset Info Service AWS Elastic Beanstalk CDN Authentication Authorisation Asset info request
- 49. CDN-aware Asset Info Service > GET /asset/31337 HTTP/1.1 < assetUrl: "http://cdn-a.alexjs.im/vod/31337.m3u8" < adProvider: "alexjsAds" < countryCode: "[im, sg, id]"
- 50. CDN-aware Asset Info Service S3 Ingest Ingest Amazon DynamoDB Media processing Amazon Elastic Transcoder Processing
- 51. CDN-aware Asset Info Service S3 Ingest Ingest AWS Elastic Beanstalk Amazon DynamoDB Media processing Amazon Elastic Transcoder Processing Authentication Authorisation Asset info request
- 52. CDN-aware Asset Info Service S3 IngestAuthoritative Source – for multi-CDN Ingest AWS Elastic Beanstalk CDN Amazon DynamoDB Media processing Amazon Elastic Transcoder Processing Authentication Authorisation Asset info request
- 53. S3 IngestAuthoritative Source – for multi-CDN Ingest API Gateway CDN Amazon DynamoDB Media processing Amazon Elastic Transcoder Processing Authentication Authorisation CDN-aware Asset Info Service Perfect for microservices – and for serverless computing Lambda
- 54. Processing & Management Content Production Content Distribution Content Storage Content Consumption
- 56. Concurrency Problem Studio mandate on stream concurrency Cross-account sharing Approach Player heartbeat Subscription-based thresholds
- 57. Subscriber Concurrency Management Heartbeat sent from every player at regular intervals Supports both native and web-based players Amazon Kinesis Lambda
- 58. Subscriber Concurrency Management { assetId: "d6f9fe" // Programme ID userId: "33114220875dc" // Token / User ID timeStamp: "T00:00:05" // Progress deviceId: "93d2d4fef95cb" // Fingerprint deviceType: "Amazon Fire" // From API } (Pseudo JSON) Amazon Kinesis Lambda
- 59. Subscriber Concurrency Management Heartbeat received by Amazon Kinesis Data fed into DynamoDB DRM licence server reads DynamoDB table Decision made Amazon Kinesis DynamoDB Lambda EC2
- 60. Subscriber Concurrency Management Amazon Kinesis Lambda Track drop-off Highlight popular content Feed back into content developmentDynamoDB Amazon Redshift
- 61. Subscriber Concurrency Management Amazon Kinesis Lambda Syndicate data to ads server Recommendations-based demographic DynamoDB Amazon Redshift Amazon MLAd server
- 62. Subscriber Concurrency Management Amazon Kinesis Lambda Cross-device Play / pause / resume No servers needed Client-side “stop” can be enabled, too (non-DRM based) DynamoDB Amazon Redshift Lambda API Gateway
- 63. Go90 — Verizon
- 65. Processing & Management Content Production Content Distribution Content Storage Content Consumption
- 66. Go90 — from 30,000 ft 100+ micro services/systems 10+ deployments per week Multiple clients, large target user base High reliability/availability, low latency, superior user experience
- 67. APIs — Edge Services
- 68. Typical Service Stack Storage Logging Service Client Monitoring Configuration Domain Logic NGINX https://github.com/oncue
- 69. Governor Accounts Log Upload Search Profile Image Discovery Event Q DRM Metadata Processing Metadata Ingest Content Ops Analytics, Ad, Curate 3rd Party Analytic s Favorites Cert Prov AWS HSM Ads Offers and Packages DAI backend STB Maestro Content Ingest CDN Reco Discovery Accounts Favorites Search DRM Profile Rights SU and…a 100 others Profiles Profiles Q
- 71. Accounts Log Upload Search Profile Image Discovery Event Q DRM Analytic s Favorites Cert Prov AWS HSM Ads Offers and Packages DAI backend STB Maestro Content Ingest CDN Reco and…a 100 others Profiles Profiles Q Governor
- 72. Accounts Log Upload Search Profile Image Discovery Event Q DRM Favorites Cert Prov AWS HSM Ads Offers and Packages DAI backend STB Maestro Content Ingest CDN Reco Discovery Accounts Favorites Search DRM Profile Rights SU and…a 100 others Profiles Profiles Q Governor Analytic s
- 73. Governor Accounts Log Upload Search Profile Image Discovery Event Q DRM Metadata Processing Metadata Ingest Content Ops Analytics, Ad, Curate 3rd Party Analytics Favorites Cert Prov AWS HSM Ads Offers and Packages DAI backend STB Maestro Content Ingest CDN Reco Discovery Accounts Favorites Search DRM Profile Rights SU and…a 100 others Profiles Profiles Q
- 74. Metadata — Ingest and Storage
- 75. Courtesy: IMDB
- 76. Nile Linear (Airings) VODLive Nile MMS (content ops) Linear Ingest Provisioning Workflow Content Types Input Process Nile Workflow Package Info Raw Data Normalized Data Blended Data Search River DB Profile Service Search River (Publish Logic) Amazon EC2 Image Processing Amazon EC2Auto Scaling Amazon S3 CDN Catalog Index Analytics River Ads River 3rd Party
- 78. Linear (Airings) VODLive MMS (content ops) Linear Ingest Provisioning Workflow Content Types Input Process Package Info Amazon EC2
- 79. Linear (Airings) VODLive Nile MMS (content ops) Linear Ingest Provisioning Workflow Content Types Input Process Nile Workflow Package Info Raw Data Normalized Data Blended Data Search River DB Amazon EC2
- 80. Nile Linear (Airings) VODLive Nile MMS (content ops) Linear Ingest Provisioning Workflow Content Types Input Process Nile Workflow Package Info Raw Data Normalized Data Blended Data Search River DB Profile Service Search River (Publish Logic) Amazon EC2 Catalog Index Amazon EC2
- 81. Nile Linear (Airings) VODLive Nile MMS (content ops) Linear Ingest Provisioning Workflow Content Types Input Process Nile Workflow Package Info Raw Data Normalized Data Blended Data Search River DB Profile Service Search River (Publish Logic) Amazon EC2 Catalog Index Analytics River Ads River 3rd Party
- 82. Nile Linear (Airings) VODLive Nile MMS (content ops) Linear Ingest Provisioning Workflow Content Types Input Process Nile Workflow Package Info Raw Data Normalized Data Blended Data Search River DB Profile Service Search River (Publish Logic) Amazon EC2 Image Processing Amazon EC2Auto Scaling Amazon S3 CDN Catalo g Index Analytics River Ads River 3rd Party
- 85. Best Practices Avoid a priori optimization Deploy often Keep simple and separate systems Don’t always give in to “right tool for the right job”
- 86. Summary
- 87. Summary • Avoid a priori optimization • Deploy often • Keep simple and separate systems • Don’t always give in to “right tool for the right job” Processing & Management Content Production Content Distribution Content Storage Content Consumption
- 88. Summary • Serverless by default • Avoid lock-in • Be cautious of the outside • Reuse data for good • Avoid a priori optimization • Deploy often • Keep simple and separate systems • Don’t always give in to “right tool for the right job” Processing & Management Content Production Content Distribution Content Storage Content Consumption
- 89. Summary • Serverless by default • Avoid lock-in • Be cautious of the outside • Reuse data for good • Avoid a priori optimization • Deploy often • Keep simple and separate systems • Don’t always give in to “right tool for the right job” Go Build.
- 90. Thank you!
- 91. Remember to complete your evaluations!