Talk outline
Bio and stuff
What does it mean when you operationalize your app?
Typical Ops feedback loop (Page 3): http://espinosa.io/control-theory/sketches.pdf
Notice how it looks the same (without the Pager!)
http://espinosa.io/control-theory/sketch2.pdf
- Compare the output to its desired reference value.
- apply change to input that counteracts deviations of the output to the reference value.
Ops example:
- you get paged for server load
- you add instances
- you go back to sleep
- desired reference value - your Ops target
- current value - reports from your monitoring system
- input - number of instances
- Stabiliy
- Performance
- Accuracy
Page 3 & 4 of http://espinosa.io/control-theory/sketches.pdf
- Proportional - error multiplied by a constant
- Integral - total error accumulated over time
- Differential - change in the error previously
Make analogies on how that looks like in an AWS Autoscaling policy
- webapp deployed on Kubernetes
- load generator
- some monitoring
- Autoscaler/ controller making Kubernetes API calls
It should look something like Out[5] in http://espinosa.io/control-theory/chapter13.html.
References:
- Janert, Feedback Control for Computer Systems http://amzn.com/1449361692
- Hellerstein et. al., Feedback Control of Computing Systems http://amzn.com/047126637X