A Scheme Reducing Task Drops for Data Dependent Tasks on Mixed Criticality Systems

Abstract

In embedded real-time systems, some tasks have varying execution times due to data dependency such as the number of input data from outside. Mixed criticality systems (MCS) can deal with these tasks by discarding non-critical tasks to obtain the computing resource when critical tasks overrun. However, there are cases in which the execution time of the task can be estimated as a form of function that has computation order depending on the algorithm inside it. Classic MCS model has difficulty dealing with this kind of task because it needs to estimate the execution time in a certain number of levels, two in most cases. We define a novel MC task model which has a worst case execution time as a function and propose a run-time task control scheme that selects the non-critical tasks which should be discarded by calculating the amount of utilization required by critical tasks overrun. We evaluate its performance in simulation and demonstrate its effectiveness.