FFOB: efficient online mode-switch procrastination in mixed-criticality systems

In mixed-criticality scheduling, the widely adopted mode-switch scheme assumes that both high- and low-criticality tasks are schedulable when no tasks overrun (normal mode) and all high-criticality tasks are schedulable even when they overrun (critical mode, where low-criticality tasks are abandoned/degraded). However, this scheme greatly impedes the system performance because triggering a mode-switch immediately after any task overruns is abrupt and pessimistic. It is urgent to solve this problem as the mode-switch scheme is a fundamental part in the mixed-criticality scheduling. In this article, we present an on-the-fly fast overrun budgeting mechanism for both, earliest-deadline-first scheduled and fixed-priority scheduled systems that can effectively keep the system away from the critical mode. Our main idea is to perform overrun budgeting for all tasks as a whole, by monitoring task executions and updating a common overrun budget. This way, the overrun budget is shared among all tasks and adaptively replenished by leveraging run-time information; consequently, mode-switch can be postponed as much as possible. Both, extensive simulations and real-life deployments demonstrate that our proposed mode-switch scheme is lightweight and significantly outperforms existing solutions in improving the system's quality of service for low-criticality tasks.