Integer Programming to optimize costs in Kubernetes

on YouTube ↗06m

" summary

Proposes an integer linear program that models pod-to-node assignment with binary open-node variables (alpha), capacity variables (U), and an error penalty (gamma) so higher pod failures raise solution cost. Implements the model in Java and solves with Gurobi via Java, compares results against the kube-scheduler implementation, and runs experiments on varying pod/node counts from the GitHub repo. Finds lower mathematical cost but longer scheduling time because the solver searches node–pod matches exhaustively.

AI-written summary · about

" related talks

• 46mIntro: Scheduling SIGDa Ma, Shivram Srivastava · KubeCon + CloudNativeCon Europe 2019
• 58mParallelism in Linear and Mixed Integer ProgrammingEd Rothberg · Parallel and Distributed Algorithms for Inference and Optimization (Simons Institute, Oct. 21–24, 2013)
• 35mSIG Scheduling Deep DiveBobby Salamat, Jonathan Basseri · KubeCon + CloudNativeCon Europe 2018
• 30mDeep Dive: Scheduling SIGBabak Salamat · KubeCon + CloudNativeCon Europe 2019
• 39mPractical Container Scheduling: Juggling Optimizations, Guarantees, and Trade-Offs at NetflixSharma Podila · MesosCon North America 2017
• 17mSolving Max-Min Fair Resource Allocations Quickly on Large GraphsUmesh Krishnaswamy, Himanshu Raj, Santiago Segarra, Pooria Namyar, Behnaz Arzani, Srikanth Kandula, Daniel Crankshaw, Ramesh Govindan · 21st USENIX Symposium on Networked Systems Design and Implementation (NSDI '24)
• 55mScaling out operational research with RustMassimiliano Mantione · RustLab 2024
• 35mApproximation hardness and online algorithms for resource schedulingRathish Das · Managing Parallelism 2025