成果蒂LFCS当前增强.pdfVIP

Agenda ● Introduction to Linux Scheduling ● Overview of the Power Aware Scheduler ● Usage of Per Entity Load Tracking Metric in Power Aware Scheduler ● Where does Power Aware Scheduler stand today? ● Experimental Evaluation of the Power Aware Scheduler ● Other Efforts towards Power Efficiency in Scheduling ● References What does the Linux scheduler do? ● The Linux scheduler does two primary jobs: - Scheduling multiple runnable jobs fairly on a CPU - Scheduling jobs optimally across multiple CPUs in an SMP system Scheduling multiple runnable jobs fairly on a CPU Requirement: The CPU bandwidth needs to be “fairly” distributed among tasks ● Data Structure used is the Red Black Tree. ● The tasks are sorted out in this tree in the reasing order of CPU bandwidth received and the leftmost task is chosen to run on the CPU. ● A new task or a woken up task is positioned at the left most end of the tree. ● The CPU bandwidth consumed by a task is called the vruntime. ● vruntime and the above data structure ensure that all tasks receive a fair share of CPU bandwidth, which in-turn ensur ter in ctivity of workloads. Goals of SMP scheduling : ● No CPU must have too many or too few tasks to run ● A new task/ woken up task must find the right CPU to run on quickly Work Conserving Nature of scheduler: New Task arrives CPU2 goes idle Load Balancing If there is CPU bandwidth, schedule T4 T4 the task on it. T1 T2 T3 T1 T2 T1 T2 T4 CPU0 CPU1 CPU2 CPU0 CPU1 CPU2 CPU0 CPU1 CPU2 CPU0 CPU1 CPU2 T4 T1 T2 T3 Queue of tasks to be scheduled SMP Scheduling