OS: Tuning Completely Fair scheduler CFS
The Linux Completely Fair scheduler (CFS) includes parameters that you can adjust to fit your specific needs.
The following parameters can be accessed through the proc file system. The default values (in nanoseconds) of these parameters from SLES 11 SP1 are shown in parentheses.
sched_min_granularity_ns (16000000): Minimum preemption granularity for processor-bound tasks. Tasks are guaranteed to run for this minimum time before they are preempted. sched_latency_ns (80000000): Period over which CFQ tries to fairly schedule the tasks on the runqueue. All of the tasks on the runqueue are guaranteed to be scheduled once within this period. sched_wakeup_granularity_ns (20000000): Ability of tasks being woken to preempt the current task. The smaller the value, the easier it is for the task to force the preemption.
To verify the value of each parameter, use the following command:
- cat /proc/sys/kernel/sched_latency_ns
To modify the value of each parameter, use the following command:
- echo 20000 > /proc/sys/kernel/sched_latency_ns
In the test environment, best results were achieved with the following values for these three parameters:
kernel.sched_min_granularity_ns = 100000 kernel.sched_wakeup_granularity_ns = 25000 kernel.sched_latency_ns = 1000000
If you want to make these changes permanent, you can modify these parameters in the /etc/sysctl.conf file and run the sysctl -p command.