On Thu, 19 Jul 2018, Johannes Thumshirn wrote: > On Wed, Jul 18, 2018 at 11:01:52AM -0400, Alan Stern wrote: > > On Wed, 18 Jul 2018, Johannes Thumshirn wrote: > > > > > On Wed, Jul 18, 2018 at 10:12:37AM -0400, Alan Stern wrote: > > > > As one extra reminder... People who care about extreme performance can > > > > perfectly well disable CONFIG_PM in their kernels. That should remove > > > > any overhead due to runtime PM. > > > > > > Yes but how likely is this for people who are running their data > > > center with a distribution kernel. > > > > Strictly speaking, that's a nonsense question. People who run > > distribution kernels are 100% unlikely to change the kernel > > configuration, by definition. > > > > That's not the point. If people really care about getting the utmost > > performance, to the extent that they resent the minimal overhead > > required for runtime PM, then they should not be running distribution > > kernels in the first place. > > Can we agree to disagree here? > > Some of the people I know in HPC, HFT, etc... run stock SLE/RHEL and > they do file bugs for every single 3-5% performance regression. This > is not something I made up, this is the kind of things I have to deal > with on my day job. The performance affect of runtime PM ought to be less than 1%. If it is larger, we'll figure out a different approach. Alan Stern
