Open Problems in Scheduling Self-Suspending Tasks

Self-suspension intervals are becoming increasingly common in various systems such as: (i) multi-core processors, where tasks running on one core have to synchronize with tasks running on other cores, (ii) heterogeneous ISA multi-core processors, where certain instructions can only be executed on specific processor cores, and (iii) systems with accelerated co-processors such as Digital Signal Processors (DSPs) or Graphics Processing Units (GPUs). In the light of these developments, a few key questions arise: (a) What should be the standard task model for specifying suspension intervals in such systems? (b) Given that classical scheduling algorithms such as Earliest-Deadline First (EDF) suffer from scheduling anomalies in self-suspending task systems, does there exist a competitive anomaly-free scheduling algorithm for such systems? (c) Given that the feasibility problem of scheduling periodic tasks with at most one self-suspension per task and implicit deadlines is NPHard, what can we say about the feasibility condition for scheduling sporadic self-suspending task systems? In this position paper, we provide some preliminary ideas and intuitions towards answering these questions, and seek to engage the broader real-time research community in solving these open problems.