Evolution of Data Infrastructure at Peak Games

Transcript

1. Evolution of data Infrastructure @
2. Serdar Sahin serdar@peakgames.net Head of Cloud & Big Data Services
3. Every 2 days we create as much information as we did from the beginning of time up to 2003.
5. Some of our GAMES
6. 70% 70% 150+ MOBILE USERS US REVENUES COUNTRIES 40% FEMALE AUDIENCE 300m+ INSTALLS 300k+ CONCURRENT PLAYERS
7. Once upon a time
8. 3,5 YEARS AGO
9. Behavioral analytics
10. Our ALTERNATIVES WERE… MPP COLUMNAR DATABASES Hadoop ecosystem In-memory
11. Why NOT Hadoop?
12. Large ecosystem Great for unstructured data Vendor agnostic Inexpensive for complex requirements
13. Hadoop & Spark Ecosystem By amplab
14. THERE ARE Other SOLUTIONS FOR ANALYSING STRUCTURED DATA
16. CONSIDER THIS QUERY
17. SQL on Hadoop execution engines & redshift Benchmark https://amplab.cs.berkeley.edu/benchmark/ (Y Axis: Seconds)
18. Our ALTERNATIVES WERE… MPP COLUMNAR DATABASES Hadoop ecosystem In-memory
19. COLUMNAR STORAGE
20. Row store vs Column store
21. RDS POSTGRES vs. redshift https://www.periscopedata.com/amazon-redshift-guide (Y Axis: Seconds)
22. CSTORE Some of column store engines/db
23. Consolidating the data sources
24. First event API
25. Value-added services
26. CRM
27. Push notifications EMAIL Facebook notifications Retargeting In-game messaging Dynamic packages
28. user_id duration_sec date 937493 937493 937493 143 663 1753 Sessions (Events) user_id score date 937493 937493 937493 937493 937493 937493 856 924 195 266 311 241 Games (Events) user_id sessions_last_30_days games_last_7_days total_session_duration 937493 2 3 2559 01-04-2016 11:45:34 20-04-2016 20:16:23 03-05-2016 17:13:05 01-04-2016 11:46:45 20-04-2016 20:17:12 20-04-2016 20:21:55 03-05-2016 17:14:23 03-05-2016 17:18:12 03-05-2016 17:25:55 Current Date: 07-05-2016 Users (CRM)
29. Segmentation
32. Stats Events / day 1b+ write / sec Event API 20k+ write / sec VoltDB 100k+
33. Why do we try so hard ?
34. What’s next?
35. We are hiring ! serdar [at] peakgames.net