Energy Efficient Assignment and Deployment of Tasks in Structurally Variable Infrastructures

Abstract

The importance of cyber-physical systems is growing very fast, being part of the Internet of Things vision. These devices generate data that could collapse the network and can not be assumed by the cloud. New technologies like Mobile Cloud Computing and Mobile Edge Computing are taking importance as solution for this issue. The idea is offloading some tasks to devices situated closer to the user device, reducing network congestion and improving applications performance (e.g., in terms of latency and energy). However, the variability of the target devices' features and processing tasks' requirements is very diverse, being difficult to decide which device is more adequate to deploy and run such processing tasks. Once decided, task offloading used to be done manually. Then, it is necessary a method to automatize the task assignation and deployment process. In this thesis we propose to model the structural variability of the deployment infrastructure and applications using feature models, on the basis of a SPL engineering process. Combining SPL methodology with Edge Computing, the deployment of applications is addressed as the derivation of a product. The data of the valid configurations is used by a task assignment framework, which determines the optimal tasks offloading solution in different network devices, and the resources of them that should be assigned to each task/user. Our solution provides the most energy and latency efficient deployment solution, accomplishing the QoS requirements of the application in the process.