What in the world is my thesis even about?
Before you even begin reading this and wondering what in the world is this guy rambling on about you need to keep in mind, this entire thesis idea came from the fact that I wanted an excuse to build a raspberry pi cluster, and it was just about finding a project I could implement on said cluster.
Currently I am writing my MSc thesis at DTU, where the main goal I am trying to achieve is to optimize Kubernetes memory usage at a cluster level via the new Swap space feature which recently entered alpha support. This would be done by identifying services which would experience little to no performance impact if part of the application memory were to be swapped into swap space.
Swap resilient services could then be moved to nodes which had swap space enabled, providing a lower overall memory footprint.
While this sounds like it would give minimal gains for individual clusters, where the real scale of this optimization comes in is when this tooling is tied into the Kubernetes virtual cluster project, theoretically this tooling could manage every single service running across an entire data center, the savings for Google or Amazon where they automatically optimized memory usage of all their client clusters would be immense… (Again keep in mind, I just wanted to build a Pi Cluster.)
Now how this identification would be done is what I am still looking at, however current ideas I am noodling with is to determine some kind of heuristic based off the Linux kernel PSI.
Now thats enough procastination, back to work.