Design and Analysis for Real-Time Mixed-Criticality SchedulingSpeaker: Arvind Easwaran – Singapore, Singapore
Topic(s): Architecture, Embedded Systems and Electronics, Robotics
Real-time mixed-criticality systems have stringent timing requirements in the form of hard deadlines and a collection of tasks having different levels of importance or criticality hosted on a single hardware platform. Avionics and automotive are two well known domains for such systems, where the criticality level has a strong correlation with the assurance levels used for certification. Traditionally, static processor partitioning, in the form of fixed allocation of processing time, has been employed to ensure isolation between the different criticality tasks and guarantee task deadlines. However, due to increasing software and hardware complexity, determining a tight bound on the worst-case execution time of tasks is becoming increasingly difficult. As a result, pessimistic upper-bounds are often used for critical tasks, and this leads to a significant processor under-utilisation when used with static partitioning. To overcome this inefficiency, the concept of mixed-criticality scheduling has emerged in the last decade. Under this paradigm, processing capacity is partitioned among all the tasks using a less conservative execution time estimate. In the eventuality that some critical task requires additional execution, the schedule is adapted to favour the critical tasks over less critical ones.
Focusing on mixed-criticality scheduling issues, in this talk I will present two recent results. Considering a single-core processor, I will present a new scheduling model and runtime budget enforcement policy to dynamically manage the budget allocations for critical tasks so that: 1) all tasks continue to receive as much budget as they need for as long as feasible, and 2) less critical tasks continue to receive some guaranteed budget even after the schedule is adapted to favour the critical tasks. Then, considering a multi-core processor, I will present a new fluid scheduling model that significantly improves the schedulability performance when compared to state-of-the-art approaches, while still having a theoretically bounded performance guarantee. Finally, I will conclude the talk with a brief overview of some of our other work in this area, and highlight some of the open problems.
About this LectureNumber of Slides: 32
Duration: 45 minutes
Languages Available: English
Request this Lecture
To request this particular lecture, please complete this online form.
Request a Tour
To request a tour with this speaker, please complete this online form.
All requests will be sent to ACM headquarters for review.