An efficient and scalable implementation of global EDF in Linux

The increasing popularity of multi-core architectures is pushing researchers and developers to consider multi-cores for executing soft and hard real-time applications. Real-Time schedulers for multi processor systems can be roughly categorized into partitioned and global schedulers. The first ones are more adequate for hard real-time embedded systems, in which applications are statically loaded at start-up and rarely change at run-time. Thanks to automatic load balancing, global schedulers may be useful in open systems, where applications can join and leave the system at any time, and for applications with highly varying workloads. Linux supports global and partitioned scheduling through its real-time scheduling class, which provides SCHED FIFO and SCHED RR fixed priority policies. Recently, the SCHED DEADLINE policy was proposed that provides Earliest Deadline First scheduling with budget control. In this paper we propose a new implementation for global EDF scheduling which uses a heap global data structure to speed-up scheduling decisions. We also compare the execution time of the main scheduling functions in the kernel for four different implementations of global scheduling, showing that our implementation is as scalable and efficient as SCHED FIFO. ∗The research leading to these results has received funding from the European Community’s Seventh Framework Programme n.248465 “S(o)OS – Service-oriented Operating Systems.”