Upgrade to Pro — share decks privately, control downloads, hide ads and more …

Observability and Control Theory

Piyush Verma
November 16, 2019

Observability and Control Theory

Piyush Verma

November 16, 2019
Tweet

More Decks by Piyush Verma

Other Decks in Programming

Transcript

  1. Observability
    & Control Theory

    View Slide

  2. Question

    View Slide

  3. Question

    View Slide

  4. Question

    View Slide

  5. Monitoring!!
    Answer

    View Slide

  6. ● You monitor what you know
    ● You react after it has failed.
    ● What about unknown-unknowns?
    ● Doesn’t avoid failures.
    ● Tells something broke and now it isn’t.
    ● Always a comparative operator.
    Question: What is Monitoring?

    View Slide

  7. ● Systems are Distributed.
    ● What about?
    ○ Accuracy
    ○ Latency
    ○ Correctness
    ○ Consistency
    ● It’s till or from Now.
    What are we missing?

    View Slide

  8. Monitoring Timeline
    Late 2000s
    ● SOA
    Early 2010
    ● Microservices
    1990s
    ● Servers
    2000s
    ● Monolith
    Now
    Lambda

    View Slide

  9. The REAL question
    If a tree falls in a forest and nobody
    heard it, did it make a sound?

    View Slide

  10. The REAL answer
    All falling trees
    yield Logs.

    View Slide

  11. Software, by default, is opaque.
    To debug & control a running
    system,
    you need observation pre-built.

    View Slide

  12. Observability

    View Slide

  13. Need
    - Debugging
    - Pattern Detection

    View Slide

  14. View Slide

  15. Unlike monitoring, observability
    is not failure-centric.
    If your job allows, you can use
    it to understand system
    performance, just-like-that.
    ¯\_(ツ)_/¯

    View Slide

  16. Stability

    View Slide

  17. Control Theory

    View Slide

  18. Control Theory

    View Slide

  19. Example

    View Slide

  20. TCP Flow Control

    View Slide

  21. TCP Flow Control

    View Slide

  22. TCP Congestion Control

    View Slide

  23. TCP Congestion Control

    View Slide

  24. Practical Applications

    View Slide

  25. ● Constrained Optimization Problems
    ● Constraint Scaling Problems
    Practical Applications

    View Slide

  26. Auto-Scaling

    View Slide

  27. Cache

    View Slide

  28. Gateway

    View Slide

  29. Load-Balancer

    View Slide

  30. Progressive Streaming

    View Slide

  31. Software, by default, is opaque.
    To debug & control a running
    system,
    you need observation pre-built.

    View Slide

  32. Thank You
    - Piyush Verma

    View Slide