Multi-core real-time scheduling in multilevel feedback queue with starvation mitigation (MLFQ-RT)

Abstract

Process scheduling for real-time processes is a critical function of real-time operating systems, which are required to guarantee soft and hard deadlines for completing real-time processes. The behavior of Multi-Level Feedback Queue (MLFQ) scheduling mechanisms intrinsically support a scheduling that favors short CPU bursts to the complete exclusion of all other processes in the ready queues. This MLFQ feature has been extended to support meeting both hard and soft real-time process deadlines in robotics and automated manufacturing applications. This research explores a new derivative of MLFQ for real-time scheduling called MLFQ-Real-Time (MLFQ-RT) investigated through simulation for multi-core processors. The MLFQ-RT real-time extension for multi-core processors builds upon research previously solved for a known weakness of MLFQ scheduling: a vulnerability to starvation of processes in the lowest priority queue, so that the operating system is unable to guarantee that all processes will make progress. This scheduling algorithm is extended to multi-core processors with three hypothesis examined and validated through simulation, showing hard and soft real-time process scheduling while maintaining the previously demonstrated mitigation of starvation in low priority queues.