Dr. Arvind Easwaran, Nanyang Technological University, Singapore
Design and Analysis for Real-Time Mixed-Criticality Scheduling
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.
Arvind Easwaran is an Assistant Professor in the School of Computer Science and Engineering at Nanyang Technological University (NTU), Singapore. He received a PhD degree from the University of Pennsylvania, USA, in 2008. Prior to joining NTU in 2013, he has been an Invited Research Scientist at the Polytechnic Institute of Porto, Portugal, and an R & D Scientist at Honeywell Aerospace, USA. He is leading several projects in NTU including real-time scheduling theory for mixed-criticality systems funded by 2 the Ministry of Education, resilient cyber-infrastructure and cyber-twin modelling for smart manufacturing funded by the National Research Foundation under Delta Electronics Corporate Lab, and design optimisation for district cooling systems funded by the Economic Development Board and Veolia City Modelling Center. He is also involved in several initiatives under the Energy Research Institute @ NTU, including as a Cluster Director for the Autonomous Vehicles program, and as a Topic Leader for Cyber-Physical Modelling in the research network on renewable energies between NTU and CNRS, France. His research interests broadly fall into the category of design and analysis of cyber-physical and real-time systems.
Invited by Professor Sebastian Fischmeister.