Big Data and Videos @ globo.com

Co-authored by @rafaelspereira and @moret. Presentation given during the EMC Summer School - http://emcbigdataschool.nce.ufrj.br/ - 2013

Transcript

1. globo .com Big data and videos Friday, February 8, 2013
2. globo .com Friday, February 8, 2013
3. what is globo.com? Friday, February 8, 2013
4. home Friday, February 8, 2013
5. Internet presence Solutions and infra-structure company Friday, February 8, 2013
6. 30 million users 45 million daily hits size (out of 50M active internet users universe in Brazil - source: Ibope) Friday, February 8, 2013
7. where are these people going? Friday, February 8, 2013
8. portals and products Friday, February 8, 2013
9. portals and products Friday, February 8, 2013
10. portals and products Friday, February 8, 2013
11. portals and products Friday, February 8, 2013
12. portals and products Friday, February 8, 2013
13. portals and products Friday, February 8, 2013
14. portals and products Friday, February 8, 2013
15. where are globo.com videos? Friday, February 8, 2013
16. videos within pages Friday, February 8, 2013
17. videos within pages Friday, February 8, 2013
18. 1 billion API requests/month 2 million videos stored 300.000 online live viewers videos within pages Friday, February 8, 2013
19. videos within pages Friday, February 8, 2013
20. videos within pages video as main content + Friday, February 8, 2013
21. globo.tv Friday, February 8, 2013
22. globo.tv Friday, February 8, 2013
23. Video encoding in reasonable time Keep viewers interested challenges Friday, February 8, 2013
24. Split & Merge Friday, February 8, 2013
25. video encoding Friday, February 8, 2013
26. video size 1080x1920, 29.97fps, RGB, 8bpp, 90min 1080 * 1920 * 29.97 * 3 * 8 * 90 * 60 / 8 = 937.62GB => 1.39Gbps Friday, February 8, 2013
27. video compression Friday, February 8, 2013
28. video compression Friday, February 8, 2013
29. speed 1 minute of video 3 minutes of encoding Friday, February 8, 2013
30. globo.com scenario Multiple devices Bandwidth from 300kbps to 10Mbps Over 3.000 hours of video monthly Friday, February 8, 2013
31. Friday, February 8, 2013
32. 200.000 online subscriptions Friday, February 8, 2013
33. process merge split map-reduce generalization Friday, February 8, 2013
34. split & merge encoding Friday, February 8, 2013
35. chunk1 chunk2 chunkN split & merge encoding Friday, February 8, 2013
36. chunk1 chunk2 chunkN chunk2 chunk1 + FULL Encoded Video chunkN + + … = split & merge encoding Friday, February 8, 2013
37. chunk1 chunk2 chunkN chunk2 chunk1 + FULL Encoded Video chunkN + + … = split & merge encoding Friday, February 8, 2013
38. EC2 Master Instance EC2 Master Instance EC2 Worker Instance EC2 Worker Instance EC2 Worker Instance EC2 Worker Instance Amazon Relational Database Service Amazon S3 (file storage) (processing) (structured data storage) Amazon SQS (job queue) aws deploy Friday, February 8, 2013
39. performance Friday, February 8, 2013
40. Globo.com EC2 S3 500 min of video / day cost Friday, February 8, 2013
41. Globo.com EC2 S3 500 min of video / day Source Video 25Mbps 91,55GB $9,15 cost Friday, February 8, 2013
42. Globo.com EC2 S3 500 min of video / day Source Video 25Mbps 91,55GB $9,15 Encoding: 2.68 min/min of video 22,3hrs $7.58 cost Friday, February 8, 2013
43. Globo.com EC2 S3 500 min of video / day Source Video 25Mbps 91,55GB $9,15 Encoding: 2.68 min/min of video 22,3hrs $7.58 Source Video Encoded Video(500kbps) 93,35GB $25.23 cost Friday, February 8, 2013
44. Globo.com EC2 S3 500 min of video / day Source Video 25Mbps 91,55GB $9,15 Encoding: 2.68 min/min of video 22,3hrs $7.58 Source Video Encoded Video(500kbps) 93,35GB $25.23 Storage 93,35GB $14 cost Friday, February 8, 2013
45. Globo.com EC2 S3 500 min of video / day Source Video 25Mbps 91,55GB $9,15 Encoding: 2.68 min/min of video 22,3hrs $7.58 Source Video Encoded Video(500kbps) 93,35GB $25.23 Storage 93,35GB $14 Encoded Video (500kbps) 1.8GB $0.30 cost Friday, February 8, 2013
46. Globo.com EC2 S3 500 min of video / day $56.26 / day $20534.9 / year cost Friday, February 8, 2013
47. split & merge Friday, February 8, 2013
48. Full coverage of the World Cup Final encoded in 2 minutes instead of 12 hours for $1 split & merge Friday, February 8, 2013
49. Nightcap Friday, February 8, 2013
50. item recommendation Friday, February 8, 2013
51. who watched this concept Friday, February 8, 2013
52. who watched this, also watched these concept Friday, February 8, 2013
53. who else does it Friday, February 8, 2013
54. who else does it Friday, February 8, 2013
55. who else does it Friday, February 8, 2013
56. model Friday, February 8, 2013
57. Friday, February 8, 2013
58. Friday, February 8, 2013
59. ? ? ? ? Friday, February 8, 2013
60. 0.5 0.5 0.25 0.25 Friday, February 8, 2013
61. calculate the similarity of 2.000.000 videos with the hits of 15.000.000 viewers in realtime scale Friday, February 8, 2013
62. 15.000.000 viewers 2.000.000 videos Friday, February 8, 2013
63. continuous map-reduce Friday, February 8, 2013
64. continuous map-reduce X X X Friday, February 8, 2013
65. continuous map-reduce Friday, February 8, 2013
66. continuous map-reduce 0.123 0.087 0.012 Friday, February 8, 2013
67. sharding #1 #2 #3 #4 #5 #6 #7 1 0 1 0 1 1 1 0 1 1 1 0 1 0 0 0 1 0 0 1 0 0.286 (2/7) Friday, February 8, 2013
68. sharding #1 #3 #5 #7 1 1 1 1 0 1 0 0 0 1 0 0 1/4 #2 #4 #6 0 0 1 1 1 1 0 0 1 1/3 (1+1)/(4+3) = 2/7 Friday, February 8, 2013
69. technology stack resque jobs and workers queues management Friday, February 8, 2013
70. technology stack in-memory key-value store Friday, February 8, 2013
71. technology stack Friday, February 8, 2013
72. globo .com Big data and videos Rafael Pereira - Danilo Moret Friday, February 8, 2013