Surviving Data in Large Doses

Transcript

1. Surviving Data in Large Doses Tareq Abedrabbo NoSQL Search Roadshow

   London 2013

2. About me • CTO at OpenCredo • Delivering large-scale data

   projects in a number of domains • Co-author of Neo4j in Action (Manning)

3. What this talk is about…

4. Supermarkets

5. Meanwhile, in DevLand

6. Bob is an application developer

7. Bob wants to build an application. Bob knows that a

   relational database is definitely not the right choice for his application

8. Bob chooses a NoSQL database because he likes it (he

   secretly thinks it’s good for his CV too).

9. Bob goes for a three-tier architecture. It’s separation of concerns.

   It’s best practice.

10. Bob builds an object model first. It’s Domain Driven Design.

    It’s best practice.

11. Bob uses an object mapping framework. Databases should be hidden

    behind layers of abstraction. It’s best practice.

12. Bob hopes for the best!

13. What challenges is Bob facing?

14. Suitability of the data model

15. Suitability of the architecture and the implementation

16. Ability to meet new requirements

17. Being able to use the selected technology to the best

    of its ability

18. Performance

19. A number of applications built on top of NoSQL technologies

    end up unfit for purpose

20. How did we get ourselves into such a mess?

21. • Technical evangelism • Evolution in requirements • Unthinking decisions

    • Ill-informed opinions

22. Common problem: there is focus on technology and implementation, not

    on real value

23. So what’s the alternative?

24. Separation of concerns based on data flow

25. Data flow

26. • Lifecycle • Structure • Size • Velocity • Purpose

27. How?

28. Identify the concerns: what do I care about?

29. Identify the locality of these concerns: where are the natural

    boundaries?

30. Build focused specialised models

31. Compose the models into a complete system

32. Computing is data structures + algorithms

33. If we accept that separation of concerns should be applied

    to algorithms, it is appropriate to apply the same thinking to data

34. The real value of this form of separation of concerns

    is true decoupling

35. What’s out there

36. CQRS

37. Polyglot Persistence

38. How do I apply it?

39. It depends on the data flow :)

40. For general-purpose data platforms, micro services work well

41. Build micro services that are closer to the natural underlying

    model

42. Other strategies are possible, for example if the data is

    highly volatile, consider in-memory grids

43. There are practical considerations - obviously

44. Don’t start with 10 different databases because you think you

    might eventually need all of them

45. How would that impact support and operations?

46. There is potential for simplification based on clearly targeted usage

47. Links • Twitter: @tareq_abedrabbo • Blog: http://www.terminalstate.net • OpenCredo: http://www.opencredo.com

    Thank you!