Reserved or On-Demand Instances? A Revenue Maximization Model for Cloud Providers

TL;DR

This paper models cloud server management to maximize revenue by balancing reserved premium and on-demand basic instances, considering costs, penalties, and energy, with an adaptive scheme based on queuing theory.

Contribution

It introduces a revenue maximization model for cloud providers managing reserved and on-demand instances, optimizing server allocation under various conditions.

Findings

The model adapts to different traffic and cost conditions.

Optimal server allocation balances penalties and energy costs.

Experimental results validate the scheme's effectiveness.

Abstract

We examine the problem of managing a server farm in a way that attempts to maximize the net revenue earned by a cloud provider by renting servers to customers according to a typical Platform-as-a-Service model. The Cloud provider offers its resources to two classes of customers: `premium' and `basic'. Premium customers pay upfront fees to reserve servers for a specified period of time (e.g. a year). Premium customers can submit jobs for their reserved servers at any time and pay a fee for the server-hours they use. The provider is liable to pay a penalty every time a `premium' job can not be executed due to lack of resources. On the other hand, `basic' customers are served on a best-effort basis, and pay a server-hour fee that may be higher than the one paid by premium customers. The provider incurs energy costs when running servers. Hence, it has an incentive to turn off idle servers. The question of how to choose the number of servers to allocate to each pool (basic and premium) is answered by analyzing a suitable queuing model and maximizing a revenue function. Experimental results show that the proposed scheme adapts to different traffic conditions, penalty levels, energy costs and usage fees.