Resource allocation and disturbance rejection in web servers using SLAs and virtualized servers

• Published in: IEEE eTransactions on network and service management Vol. 6; no. 4; pp. 226 - 239
• Main Authors: Kjaer, M.A., Kihl, M., Robertsson, A.
• Format: Journal Article
• Language: English
• Published: New York IEEE 01.12.2009; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Changing environments; Control Engineering; Control systems; Costs; Discrete event simulation; disturbance rejection; Electrical Engineering, Electronic Engineering, Information Engineering; Elektroteknik och elektronik; Engineering and Technology; feed-forward; Feedback; Feedforward systems; Networks; On-line systems; Online; online estimation; Parameter estimation; prediction; Reglerteknik; Resource management; Resource virtualization; response-time control; Servers; Studies; Teknik; Testing; virtualization; Web server; Websites
• ISSN: 1932-4537; 1932-4537
• DOI: 10.1109/TNSM.2009.04.090403

Resource management in IT-enterprises gain more and more attention due to high operation costs. For instance, web sites are subject to very changing traffic-loads over the year, over the day, or even over the minute. Online adaption to the changing environment is one way to reduce losses in the operation. Control systems based on feedback provide methods for such adaption, but is in nature slow, since changes in the environment has to propagate through the system before being compensated. Therefore, feed-forward systems can be introduced that has shown to improve the transient performance. However, earlier proposed feed-forward systems have been based on offline estimation. In this article we show that off-line estimations can be problematic in online applications. Therefore, we propose a method where parameters are estimated online, and thus also adapts to the changing environment. We compare our solution to two other control strategies proposed in the literature, which are based on off-line estimation of certain parameters. We evaluate the controllers with both discrete-event simulations and experiments in our testbed. The investigations show the strength of our proposed control system.