2731. Beyond Optimality: Pragmatic AWS Instance Placement with Hexaly Solver
Invited abstract in session MD-38: (Deep) Reinforcement Learning for Combinatorial Optimization, stream Data Science meets Optimization.
Monday, 14:30-16:00Room: Michael Sadler LG19
Authors (first author is the speaker)
| 1. | Rubén Ruiz
|
| Departamento de Estadistica e Investigación Operativa Aplicadas y Calidad, Universitat Politècnica de València |
Abstract
This paper presents a practical approach to solving the AWS EC2 virtual machine placement problem. We model the problem as a multi-dimensional, multi-objective, heterogeneous vector bin packing Mixed-Integer Linear Program (MILP). The model incorporates all relevant real-world EC2 features. The sheer size of the problem presents serious challenges to commercial mathematical solvers. While optimality is often desired in Operations Research, we need to be pragmatic in the sense that models are, after all, approximations of reality that often rely on estimated input data. We present a SET-based formulation based on Hexaly (formerly LocalSolver) solver. We compare two state-of-the-art mathematical solvers, namely Xpress and Gurobi, with Hexaly, demonstrating workable solutions where mathematical solvers go completely out of scope. Results indicate improvements over existing production placement algorithms. Our findings suggest that this practical modeling approach, which embraces the reality of working with approximate data and accepts near-optimal solutions, provides valuable insights to guide future EC2 instance placement algorithm development.
Keywords
- Large Scale Optimization
- Practice of OR
Status: accepted
Back to the list of papers