On 2015-11-17 15:10, Adam Jackson wrote: > On Wed, 2015-11-11 at 22:02 -0800, Keith Packard wrote: >> This allows the server to call GetTimeInMillis() after each request is >> processed to avoid needing setitimer. -dumbSched now turns off the >> setitimer. > > I'm not sure there are real systems we'd be protecting by not requiring > setitimer. But, at least on my Ivybridge, this only dings noop > performance by like 1% when using -dumbSched. Probably this would be > worse on non-vdso setups, but anyone with both of those problems is > already in pretty bad shape.
Data point: I've got a Dell box (also Ivy Bridge) where calling GetTimeInMillis() after each request dings noop performance by like 30% because the TSC is unstable. Performance would be fine on Fedora, where the server uses CLOCK_MONOTONIC_COARSE, but Ubuntu's CLOCK_MONOTONIC_COARSE only has a 4ms resolution so it falls back to reading the HPET every time. That said, this box is an outlier. Every other recent box I can put my hands on has a fast GetTimeInMillis(). Peter Harris -- Open Text Connectivity Solutions Group Peter Harris http://connectivity.opentext.com/ Research and Development Phone: +1 905 762 6001 phar...@opentext.com Toll Free: 1 877 359 4866 _______________________________________________ xorg-devel@lists.x.org: X.Org development Archives: http://lists.x.org/archives/xorg-devel Info: http://lists.x.org/mailman/listinfo/xorg-devel