Adjusting Real-Time Mode Transitions via Genetic Algorithms
Gordon Chalmers, Shelby H. Funk · 2017
The problem of mode transition schedulability of real-time systems is examined using a genetic algorithm. Due to the mode transition, there can be an overuse of the resource, i.e. the processor(s). A mode transition deadline is a time period in which the outgoing mode(s) have to finish, and then after the mode transition the resource can be completely used by the incoming mode(s). The primary goal of this work is to formulate sensitivity analysis of mode transitions using a genetic algorithm, such that the task set parameters are changed as little as possible to meet the mode transition deadline. This work increases the periods of incoming tasks to ensure that mode change deadlines are met. When the mode change request occurs, we have conflicting goals - start the new mode and finish the mode transition as soon as possible. If new mode tasks need to start before old mode tasks are able to complete, we need to take some corrective action. Some possibilities include : aborting old mode tasks, extending periods of new mode tasks, or reducing execution time of either old mode or new mode tasks. Our research explores the approach of increasing new mode task periods. Earlier works used task-by-task algorithms. This is unlike the genetic algorithm approach which treats the entire task set at once. The genetic algorithm can schedule the tasks such that the mode transition deadline is met sooner than the previous techniques and with higher utilizations. Deadline increases of incoming task periods can lessen the mode transition time and improve the utilization of the resource. Rather than task by task changes of previous works, the genetic algorithm can find a global change which is smaller.