This document discusses using machine learning and artificial intelligence for workload analytics and placement within a cloud environment. Specifically:
- The author is researching using machine learning for infrastructure architecture design and experimenting with crowd-sourced telemetry data for trend analysis, pattern matching, and predictive modeling.
- The problem statement asks if cloud architecture design can be optimized using crowd-sourced system data.
- The document provides an overview of using a machine learning engine to develop data models and visualize data to recommend optimized solutions for sizing and placement of workloads in the cloud.
4. Virtualisation does optimise infrastructure…
Virtualisation tool == production environment
Data from virtualisation tool usage == design environment optimisation
5. The Goldilocks Principle
Learning and understanding our customer base
• Goldilocks asked for porridge
– What does that really mean?
– Was temperature part of her request
• We’ve learned other details from this experiment
– Similarities between goldilocks and baby bear
– We know what daddy, mammy and baby bear like
• Now, Can we provide porridge for a new
character, e.g. uncle bear
– What has he asked for?
– Who is his closest match?
Too Hot Too Cold Just RightJust Right Just Right Just Right
12. Ideal Region
NE quadrant
AI search optimisation
Best Fit
Better FitGood Fit
Cost
Performance
• Normalisation (specification):
Specification attributes calculated as a
represented value of the specification
• Normalisation (directional):
- maximum values = no change
- minimum values = inverse power (-1)
• State space:
Data points represent all potentially
viable solutions
• Ideal model:
Approximates North-East trajectory
• Model heuristic navigation:
- maximum magnitude
- minimum distance from ideal
%
% -1
13. Ideal Region in
NE quadrant
Modeling complexity
Best
Fit
Better
Fit
Good
Fit
Cost
Performance
16. Machine Learning (AI) for workload analytics
and placement within a cloud environment
Kenneth Moore
Editor's Notes
There was important additional information missing in her request --- Temperature
Goldilocks asked for porridge
Did we have the full specification to fulfil her request?
Her temperature specification is also required
We’ve learned other details from this experiment
We learned the parameters of porridge required for three other customers
daddy bear, mommy bear and baby bear
Now, Can we provide porridge for a new character, e.g. uncle bear
What has he asked for?
Who is his closest match?