The document discusses Leaseweb's implementation of AI in optimizing virtual machine (VM) placements within their cloud infrastructure, covering platforms in multiple countries. It details various machine learning approaches such as supervised, unsupervised, and reinforcement learning used to enhance resource allocation, reduce costs, and analyze system states. Additionally, it highlights the use of fuzzy functions for cloudstack status evaluation and provides insights into the coding aspects of the project.

1. Leaseweb CloudStack
Using AI to place VMs

2. Sina Kashipazha
• Software Engineer @Leaseweb
• Love Java and Cloud
• Hobbies:
• Cooking
• Coding
• Scale models

3. • Virtual Private Servers
• Private Cloud Resource pools
• Leaseweb Internal
Globally operating Infrastructure as
a Service provider. Active
CloudStack platforms in:
• Germany
• Netherlands
• Singapore
• United Kingdom
• United States

4. 1. Better use of
resources
Targets
3. Save costs
2. Ability to compare
the placements

5. The How
How we approached this initiative and what we have learned along the way

6. Supervised Learning
• When input and output, both labels are known
• model learns from data to predict output for similar input data
Unsupervised Learning
• When output data is unknown
• it is needed to find patterns in data given
Reinforcement Learning
• Algorithms learn to perform an action from experience.
• Here algorithms learn through trial and error, which action yields the greatest rewards.
• The objective is to choose actions that maximize the expected reward over a given amount of time.
Which AI?

7. Simulation
Elements
• Simulation of hosts, VMs, Service Offerings
Functionalities
• Only required functionality for VM placement
Evaluation
• Way of comparing the state of the system
Fuzzy
• Fuzzy functions to describe the Cloudstack Status

8. Let’s see the code J