The Maths of Microscaling

Using control theory to scale containers in real time, in response to demand

Transcript

1. The Maths of Microscaling Liz Rice @lizrice | @microscaling
2. What is Microscaling? Assumptions Some theory Some experiments
3. What is Microscaling?
4. Traffic spike
5. Too much work Spare capacity
6. container scaling work performance metrics
7. work performance metrics container scaling VM autoscaling
8. True for regular autoscaling too VMs take much longer to scale
9. Orchestration Heterogenous services Cattle not pets
10. Performance targets
11. How many containers? Request processing time Rate of requests known? predictable?
12. performance target actual performance error time t
13. performance target p time t actual performance x e(t) = x(t) - p(t) e(t) → 0 error e
14. x(t) is proportional to n(t) n(t) = k x(t) error e time t number of containers n
15. x(t) is proportional to n(t) nope! error e time t number of containers n d(t) is proportional to e(t) d
16. Time delays It’s a dynamical system
17. Woah, the future! error e time t d(t) is proportional to e(t + T) T d
19. Control theory!
20. PID controller
21. error e time t Proportional term d(t) = Kp e(t) The further we are below target the more containers we need
22. error e time t Derivative term The faster we approach target the fewer containers we need d(t) = Kp e(t) + Kd ė(t)
23. error e time t Integral term d(t) = Kp e(t) + Kd ė(t) + Ki e(t) Offset errors accumulated over time ∫
24. Which values for K? Discrete containers?
25. Simulator goo.gl/KAqT5y
26. It works! But it’s non-trivial to tune
27. Known behaviours Machine learning
28. Container parameters = metadata microbadger.com
29. github.com/microscaling @lizrice | @microscaling app.microscaling.com microbadger.com