Hardware-based schedulers approaches for Linux OS

Tasks scheduling is one of the most critical activities in any operating system. Choosing which task should possess a CPU core (and for how long) heavily influences the system's performance. A common approach for solving these problems in today's scheduling algorithms is to consider the task's priority and execution time of all processes. However, modern hardware optimization structures, such as hierarchical caches and branch prediction units, influence the execution time, leading to suboptimal results if misused. Even though such information is made available to software through Performance Monitoring Counters (PMCs), no known scheduler uses them to improve scheduling decisions. This work integrates PMC-based evaluations in the Linux Completely Fair Scheduler (CFS) to study how thread scheduling may benefit from such knowledge. Results confirm the broad fluctuations in task turnaround time depending on the monitored structures, providing insights into possible CFS optimization.