CSR: Small: Collaborative Research:Exploiting Predictability & Interdependency of Physical Parameters for Resource-Efficient Integration of Real-Time Embedded Systems

Wayne State University, Detroit MI

Investigators

Abstract

Everyday activities are increasingly enabled by embedded systems. In such systems, a deadline violation can lead to dangerous and/or disastrous consequences. The timing parameters of such systems are driven by the physical state of these systems (e.g., ignition timing in automotive systems is determined by engine speed). Current approaches to deal with the dependencies between the physical environment state and timing parameters make very simplistic assumptions, resulting in wasted hardware and energy resources. This research facilitates more effective integration of embedded systems --thus enabling smaller technology via incorporation of techniques from mechanical engineering, control systems, and real-time scheduling. The resultant techniques will minimize the carbon footprint of systems that are ubiquitous and run our everyday lives. The educational objective of this project incorporates resulting artifacts into undergraduate- and graduate-level courses, as well as in K-12 outreach at local schools via summer camps and challenge sessions. This research aims to minimize the computational utilization of real-time embedded systems by analyzing, modeling, and predicting dependencies between the physical state and timing parameters to reduce the size, weight, and power requirements of next-generation embedded systems. A central objective is development of run-time scheduling algorithms and associated schedulability analysis for new real-time task models that characterize multiple physical dimensions of the system and their interdependencies. Run-time adaptation protocols effectively allocate resources when the estimates of the physical parameters are inaccurate. Implementations of the scheduling and allocation algorithms upon embedded platforms provide evaluation of performance of multi-dimensional techniques compared to the traditional single-dimensional models.