Online Learning for Function Placement in Serverless Computing

TL;DR

This paper introduces a novel online learning algorithm based on multi-armed bandits for optimal function placement in serverless computing, demonstrating rapid learning, good practical performance, and scalability in large networks.

Contribution

The paper presents a new online algorithm for function placement using multi-armed bandits, with proven regret bounds and an acceleration technique for large-scale networks.

Findings

Algorithm learns optimal placement rapidly

Achieves low regret growth rate of O(N M √(T ln T))

Performs well in large networks with limited computational resources

Abstract

We study the placement of virtual functions aimed at minimizing the cost. We propose a novel algorithm, using ideas based on multi-armed bandits. We prove that these algorithms learn the optimal placement policy rapidly, and their regret grows at a rate at most $O( N M \sqrt{T\ln T} )$ while respecting the feasibility constraints with high probability, where $T$ is total time slots, $M$ is the number of classes of function and $N$ is the number of computation nodes. We show through numerical experiments that the proposed algorithm both has good practical performance and modest computational complexity. We propose an acceleration technique that allows the algorithm to achieve good performance also in large networks where computational power is limited. Our experiments are fully reproducible, and the code is publicly available.