Project title / Introduction /Goal / Availablity / Prerequisites / Size / Targeted students

Title of the project:

Finite Horizon Prediction of Computer System State and its Applications to Reconfigurable Component-Based Real-time Systems

Introduction:

In a reconfigurable component-based real-time system, the system is reconfigured and modified during  run-time by adding, changing, or removing software components. This implies that a system evolves and the operation of the system is tailored according to the needs of the environment with which the system is working. However, modifying the software of the system by adding or removing components results in changes in the system state. For example, the utilization, quality, or the deadline miss ratio may change due to longer execution times. This means that a computer system becomes time-variant as the behavior of computer system varies over time.

Feedback control has been introduced as a promising foundation for performance control of real-time systems. Using the reference, overshoot, and settling time, one can explicitly specify the desired steady-state and transient-state performance of the system. The feedback structure then makes sure that the specification is guaranteed. However,  classical control theory relies on time-invariant systems and, hence, the feedback controllers do not conform to changes in the controlled system. To improve the performance of feedback controllers, one can predict how the system state is to evolve due to a change (e.g., when the user of the system reconfigures the system) in the system. This means that the prediction of the system state over a finite horizon must be determined. For example, we can predict how the utilization is to change when the system is reconfigured. Using this prediction, one can improve the control performance by feedforwarding the prediction. (Feedforwarding of state prediction is not part of this project, see project proposal below.)

So far, we have developed method for reconfiguration of component-based real-time systems and ensuring the system performance using time-invariant feedback control techniques. In this project we are interested in determining discrete time models (based on difference equations) which are used to predict the state of the system due to a reconfiguration of the system. The prediction is then used to determine whether the predicted system state is acceptable and a reconfiguration can be carried out. Furthermore, we would like to determine necessary information that would ensure that a component, developed by a third party, would provide enough information so it can be used for predicting the state of the system. Having this information stored in a prediction interface of a component would ensure component reusability across different real-time systems. This project proposal has been developed together with representatives from the industry.

Project Goal:

The goal of this project is to predict the change of the system state, induced by a reconfiguration, and extract the prediction interface of a reconfigurable component.

Status:

Available

Prerequisites:

Basic knowledge in control theory

Size:

20p

Start:

As soon as possible

Target Students:

Y, D, and IT