On 2017/7/10 16:46, Peter Zijlstra wrote: > On Mon, Jul 10, 2017 at 09:38:30AM +0800, Aubrey Li wrote: >> We measured 3%~5% improvemnt in disk IO workload, and 8%~20% improvement in >> network workload. > > Argh, what a mess :/ > > So how much of the gain is simply due to skipping NOHZ?
netperf reports 6~7% improvement due to skipping NOHZ, but still have 2~3% improvement due to excluding RCU idle enter/exit, c-state selection and other deferrable items. Also a shallow HW c-state is used in the idle loop, which means faster HW sleep enter and exit. > Mike used to carry a patch that would throttle NOHZ. And that is a _far_ small > and simpler patch to do. > Can you please point me to the link of Mike's patch? Thanks, -Aubrey