Dynamic adjustment of energy consumption for real-time tasks by integrating feedback planning and multi-frequency control

Abstract

The energy consumption of the autonomous real-time control embedded systems is a subject of openness at a technological level, it affects the operation time due to the use of batteries for its feeding, and it affects the temperature, causing undesired behaviors.
A multi-frequency control strategy is designed to regulate a computational resource, such as the processor, by dynamically assigning the computation time, globally, to all instances of a task, or locally, to each instance of  execution, adding inter- and intra-attribute adjustment characteristics, with the intention of scaling the speed of the processor, and therefore, the energy consumption. It resorts to planning fed back with energy savings, which uses control theory to plan the resources of a computer system. As well as, the guidelines of the multi-frequency control, the dynamic scaling of voltage/frequency, and the techniques for the utilization of the dynamic idle time. The real time tasks considered are periodically critical. A set of tasks for comparative tests
was taken to validate the operation, varying the computation times consumed. The global inter-task and local inter/ intrathora behaviour, give an energy consumption from 44.70% to 90.00 %, and 42.00 % to 86.66%. It is concluded that the assignment of the computation times is operated in a natural way by the multi-frequency control loop, without violating the time restrictions, providing effects in the energy consumption by the tasks.