Tetris in d dimensions: how we rightsize Kubernetes nodes
Summary
The article explains node rightsizing in Kubernetes as multi-dimensional vector bin packing. It shows why one-dimensional (size) sorting fails and introduces bin-centric heuristics (FFDProd, FFD AvgSum, Dot Product, L2) to place pods by remaining space across multiple resources. It also covers how to size pods using P95-based rightsizing, how production pipelines validate placements, and what the approach does and does not guarantee.