Feedback controllers in the cloud
Tharindu Patikirikorala, Alan Colman · Swinburne figshare (Swinburne University of Technology) · 2010
Autonomic management of quality of service attributes by dynamic resource allocation is one of the requirements in cloud computing environments. In order to provide these requirements in a flexible way existing cloud providers expose static rule /threshold/heuristic based decision implementation frameworks to their consumers. These rule /threshold/heuristic based methods are relatively easy to design and develop. However, they suffer from lack of well-founded design process to decide important design parameters (e.g.: thresholds, the number of instances to add/remove, calm time) and difficulty of dynamically adjusting to different conditions. In addition, incorrect/non-adaptable rule settings could cause long term instabilities leading to service outages. The feedback control has been shown to be useful for performance management and resource allocation in many complex software systems. In this work, we investigate the advantages and limitations of applying feedback controllers in cloud platforms. In addition, we illustrate the suitability of standard feedback controllers depending on the consumer requirements/applications. Finally, we propose a novel platform as a service architecture to design, develop, integrate and runtime manage feedback controllers for the cloud consumer applications.