A control-theoretical approach to thread scheduling for multicore processors

Feedback control has been applied to computing systems, usually taking a designed system and closing a loop to adjust some of its parameters. However, the design of computing systems components as controllers have shown advantages with respect to state-of-the-art techniques, especially in the scheduling domain, where uniprocessor schedulers have been designed as discrete-time control structures. However, the most recent computing devices (from smartphone to personal computers) have more than one core and the devised techniques cannot be applied to this context directly. This paper provides the necessary foundation to address the multicore scheduling problem as a control problem, as an extension of the uniprocessor case. We qualify the quantities to be measured and used as feedback signals for tackling the extension. We also present some control solutions and compare them using a simulator, publicly available to foster the research on the topic. The comparison shows that the devised policies have low computational complexity but achieves very good results in terms of scalability.