Anticipatory Configuration of Resource-aware Applications

Vahe Poladian, Joao Pedro Sousa, Frank Padberg, and Mary Shaw

Proceedings of the 7th International Workshop on Economics Driven Software Engineering Research, affiliated with the 27th International Conference on Software Engineering, St. Louis, MO, May, 2005.

Online links: PDF

Abstract
We propose an improved approach to dynamic configuration of resource-aware applications. The new anticipatory model of configuration maximizes utility based on three inputs: user preferences, application capability profiles, and resource availability. In this respect, the proposed model is similar to a model of configuration described in [2]. However, the latter addresses the dynamic nature of the problem by reacting to changes (such as decrease in resource availability), and maximizes the utility in a point-wise manner. The newly proposed anticipatory approach explicitly models the duration of the task and leverages possible information about the future (such as stochastic resource availability over the expected duration of the task). We expect that the anticipatory model will improve user's utility, conserve scarce resources, and reduce the amount of disruption to the user resulting from changes when compared to the reactive model. However, the optimization problem underlying the anticipatory model is computationally more difficult than the problem underlying the reactive model. We would like to investigate if the anticipatory approach is feasible and efficient in practice while delivering the above-mentioned improvements. In this paper, we carefully state the model of anticipatory configuration, highlight the sources of complexity in the problem, propose an algorithm to the anticipatory configuration problem, and provide a roadmap for research.
Keywords
multi-dimensional utility, resource, stochastic process, dynamic configuration, engineering for value


For further information, please visit the home pages of the ABLE research project and Carnegie Mellon University's Composable Systems Group.

Last modified: 5/9/2005. For comments and problems, contact able-help@cs.cmu.edu.