• linux
• kernel
• real-time scheduling
• partitioned scheduling

The work developed for this thesis has involved the design and implementation of variants of the SCHED_DEADLINE scheduler for real-time tasks in the mainline Linux kernel, with the objective of comparing different adaptive partitioning strategies. Specifically, the partitioning heuristics First-Fit and Worst-Fit have been realized as in-kernel modifications to the SCHED_DEADLINE code. These have been extensively evaluated and compared with the performance of the current global-EDF algorithm already present in SCHED_DEADLINE. The evaluation has taken into account several tasksets deployed in a multi-core system. The tasksets have been randomly generated so that their overall utilization fits in a specific configuration of the system with a given number of cores. The gathered results are discussed in depth considering the various configurations of the system.