Predictive autoscaling on K8s

Question

I am learning about using Machine Learning for provisioning proactive resources on the Kubernetes and I have found some of the interesting articles about it on the internet(for instance for Thoth or Microscaler). I have also read about predictive autoscaling in AWS and GCP however only for VM's, not to use in the K8s clusters.

https://docs.aws.amazon.com/autoscaling/plans/userguide/how-it-works.html https://cloud.google.com/compute/docs/autoscaler/predictive-autoscaling#how_predictive_autoscaling_works

However I would like to understand more in-depth about the above solutions - how do they work internally and what is the possibility of moving that solutions to the K8s. Do You know where can I find informations about it? Or if that is even possible to find those information or they are hidden internally in those companies? Do You have any other scientific articles or solutions which are now available?

Thanks

Answer 1

These are presumably proprietary tools for EC2 and GCE respectively. The need for predictive VM autoscaling is pretty strong for Virtual Machines since provisioning time can be measured in the 10's of minutes depending on the configuration.

Conversely, application spin-up time in Kubernetes is typically measured in seconds. Predictive scaling for the Kubernetes HPA might have limited practical benefit.

Now using cloud provider predictive autoscaling to scale out the VMs running a Kubernetes cluster based upon ML could be pretty interesting.