Implementation in Networked and Embedded Systems

Many systems rely on networked and embedded systems for both user equipment and network provisioning and control. The systems may have complex software structures and distributed network functionalities and the efficient and scalable use of physical resources is becoming one of the most important design aspects. Resource-aware design requires adequate methods for allocating physical resources both statically and dynamically. There is an increased focus on dynamic approaches, i.e., adaptive resource management. Increasingly common is also more holistic, or cyber-physical, approaches that also take the interaction between the computing systems and its environment into account.

This area is aimed at advancing the state of the art in networked and embedded systems, by combining the scientific expertise from the telecommunication and the control communities at Lund University. Important components in this research field are queuing theory, distributed control, real-time systems, non-linear control and estimation theory. The main goals are to develop reliable performance estimation methods and to design and optimize resource-aware control mechanisms for networked and embedded systems.


Cloud control

Cloud control is based on a recently initiated collaboration with the Cloud research group at Umeå University. We propose a control theoretic approach to a range of cloud management problems, aiming to transform today´s static and energy consuming cloud data centers into self-managed, dynamic, and dependable infrastructures, constantly delivering expected quality of service with acceptable operation costs and carbon footprint for large-scale services with varying capacity demands. Such data centers will form the backbone of the digitalized society by providing unparalleled information storage and processing capabilities. We will develop and validate performance models that accurately capture the dominant control dynamics. Long-term prediction of capacity requirements is used for admission control. A stochastic elasticity controller predicts the application´s required capacity using performance feedback. The feedback-based placement controller maps capacity to virtual machines and determines their physical placement, coordinated with a data center-wide analysis of energy consumption and temperature distribution. A holistic manager monitors the overall system behavior and steers the controllers´ concerted actions. For all problems, we investigate methods for near-optimal solutions in sufficient time, and evaluate the solutions analytically, by simulations and by large-scale experiments in our testbeds.

Adaptive Resource Management for Embedded Systems

Embedded systems are becoming increasingly complex. At the same time, the components that make up the systems grow more uncertain in their properties. For example, current developments in CPU design focuses on optimizing for average performance rather than better worst case performance. This, combined with presence of 3rd party software components with unknown properties, makes resource management using prior knowledge less and less feasible. Traditionally, embedded computing systems have been designed using static worst case assumptions on availability of resources, but for systems with large variability in use cases or which are executing on uncertain executing platforms this is increasingly difficult. An alternative approach is to instead allocate resources to different applications or systems dynamically, based on measurements of resource consumption, resource availability and the generated quality of service. This is also referred to as adaptive resource management and is particularly suited for applications of a soft real-time nature, e.g., media processing applications. However, many control and service robotic applications also fall within this domain.

Event-Based Control

We aim to develop theory and design methodology for event-based state estimation and control over networks. Working in a stochastic control setting, the goal is to reduce the state variance and/or the number of network transmissions. This is accomplished by communicating and controlling only when there is a significant deviation in the system. We will develop suboptimal estimation and control schemes with performance bounds and compare the results with the linear time-invariant case. While the theory is developed for general linear plants, we plan to apply some of the results to server systems.

Control of coordinated multipoint communication systems using localization based prediction

Significant gains in spectrum efficiency and communication bandwidths can be achieved with so called coordinated multipoint transmission. Future developments of the LTE system use a central single node that controls the downlink transmission from a multitude of transmission stations to a multitude of users, creating a large-scale MIMO system spanning several hundreds of meters and 10-100 simultaneous receivers. In such a system channel estimation and prediction is a key element. This project builds on existing work in the field but adds the use of localization-based information from GPS combined with inertial sensors in the mobile receivers to improve the performance.