Schedulability analysis for a mode transition in real-time multi-core systems

To enable real-time systems to adapt to dynamically changing environments, update functionalities and/or accommodate those tasks migrated from other failed sub-systems, there have been a number of studies on making timing guarantees while accounting for change of parameters and addition/deletion of tasks. While most of them have dealt with 'transition' protocols that delay next task releases or discard the unfinished tasks released before the transition, such protocols are not suitable for many control systems in which missing/delaying control updates (by completing periodic tasks) even during a transition or mode-change may cause system instability or incur a significant incremental operational cost. In this paper, we focus on a transition protocol that does not miss/delay control updates during a system transition, and develop a new schedulability analysis for the transition in a real-time multi-core system, which provides sufficient timing guarantees without requiring any online information, such as the release and execution patterns of tasks and the start time of a transition. To achieve this, we extend an existing popular schedulability analysis framework for nontransitional tasks, and identify the scenarios that maximize the duration of a task's interference to another task in the case of a transition. Since the analysis works for any arbitrary transition order of tasks, we can improve the schedulability performance by enforcing a specific order. We formulate the problem of assigning an optimal transition order, and develop a solution by deriving some properties of optimality. Our evaluation results demonstrate that the proposed solution finds more schedulable task sets, which are not covered by naive approaches.