Decentralized LPV-MPC controller with heuristic load balancing for a private cloud hosted application. (July 2020)

• Record Type: Journal Article
• Authors: Singh, Durgesh; P.S., Saikrishna; Pasumarthy, Ramkrishna; Krishnamurthy, Diwakar
• Is Part Of: Control engineering practice. Volume 100(2020)
• Publication Date: 2020-07
• Subjects: Cloud computing -- Web-servers -- Linear parameter-varying systems -- System identification -- Virtual machine scheduling -- Model predictive control; Automatic control -- Periodicals; 629.89
• DOI: 10.1016/j.conengprac.2020.104438
• Languages: English
• ISSNs: 0967-0661

Abstract

Abstract: Web-services are increasingly being deployed on cloud platforms that utilize virtual machines (VMs) to share physical resources among many subscribers. This allows a service provider to leverage features such as on-demand resource provisioning with a pay-as-you-go pricing model. An important consideration for the provider of a web-service is the response time experienced by the end-user. This work relies on control theory for performance guarantees of web-servers hosted on the cloud. The objective is to instantiate the optimal number of VMs to keep the response time, below a certain threshold. Firstly, a linear parameter-varying (LPV) model is developed for each of the hosted server. Secondly, a decentralized LPV-MPC controller with workload prediction, is designed to ensure performance under varying workload. The incoming workload distribution across different servers is performed based on a heuristic algorithm consisting of three modes: regular, surge, and energy optimization. The results are validated on an experimental test bed. Highlights: Identification of a global LPV model for a web-server system. Web-server workload prediction using time-series analysis. MPC with workload prediction. Development of decentralized LPV-MPC controller. Heuristic load balancing across web-servers.