From Competitive to Cooperative Resource Management for Cyber-Physical Systems

Sammanfattning: This thesis presents models and methods for feedback-based resource management for cyber-physical systems. Common for the scenarios considered are severe resource constraints, uncertain and time-varying conditions and the goal of enabling flexibility in systems design rather than restricting it. A brief survey on reservation-based scheduling, an important enabling technology for this thesis, is provided and shows how modern day resource reservation techniques are derived from their real-time system and telecommunications theory roots. Techniques for modeling components of cyber-physical systems, including both computational and physical resources, are presented. The cyclic component model, specifically designed to model common resource demanding components in smart phones, is introduced together with techniques for model parameter estimation. The topic of competitive resource management, where the different parts of the system compete for resources, is discussed using a smart phone platform as motivating example. The cyclic component model is used to form a rate-based performance metric that results in a convex optimization problem. A specialized optimization algorithm for solving this problem efficiently online and with limited precision hardware is introduced and evaluated through simulations. A feedback control scheme for distributing resources in cases where components collaborate, i.e., where the performance metric is dependent on more than one component, is detailed and examined in a scenario where the available resource is limited by the thermal dynamics of the CPU. The scheme is evaluated through simulation of a conversational video pipeline. The thermal model is validated on a mobile robot, where it is used as part of an adaptive resource manager. The problem of energy conservative distribution of content to a population of co-located mobile clients is used to motivate the chapter on cooperative resource management, i.e., scenarios where the participants have individual but similar goals and can benefit from sharing their partial results so that all collaborators save cost. The model for content trading is presented in synchronous and asynchronous formulations and performance is evaluated through both simulations and experimental results using a prototype implementation in an emulated environment.