Response-Time-Optimized Distributed Cloud Resource Allocation

TL;DR

This paper addresses the challenge of optimally assigning users to distributed cloud resources to minimize response times, considering server utilization and queuing delays, with a focus on practical, scalable solutions.

Contribution

It introduces a convex capacitated facility location model with integrated queuing systems and a linear reformulation for faster solutions, enhancing response time optimization in distributed clouds.

Findings

Linear reformulation shows small optimality gaps and faster solving times.

Considering queuing delays significantly reduces response times.

The model effectively balances resource costs and service quality.

Abstract

A current trend in networking and cloud computing is to provide compute resources over widely dispersed places exemplified by initiatives like Network Function Virtualisation. This paves the way for a widespread service deployment and can improve service quality; a nearby server can reduce the user-perceived response times. But always using the nearest server is a bad decision if that server is already highly utilized. This paper investigates the optimal assignment of users to widespread resources -- a convex capacitated facility location problem with integrated queuing systems. We determine the response times depending on the number of used resources. This enables service providers to balance between resource costs and the corresponding service quality. We also present a linear problem reformulation showing small optimality gaps and faster solving times; this speed-up enables a swift reaction to demand changes. Finally, we compare solutions by either considering or ignoring queuing systems and discuss the response time reduction by using the more complex model. Our investigations are backed by large-scale numerical evaluations.