Re: [Ntp] Antw: Re: Antw: [EXT] Minutes from IETF 108 NTP WG session

[Miroslav Lichvar <[e-mail]>]

On Mon, Aug 03, 2020 at 05:12:41AM -0700, Hal Murray wrote:
> >> (D)get_res: smallest delta is 0.000001514
> >> (D)get_res: largest delta is 0.000001695
> 
> > This suggests the system is not using the TSC clocksource, but something
> > slower like HPET. Check the following files: 
> 
> Maybe, but I'll bet a bigger contribution is doing a syscall rather than using 
> vsdo.

That depends on hardware and kernel configuration. Looking at some of
my old tests, the overhead of the syscall was much smaller then
reading of the HPET itself. However, on Intel CPUs some of the
security mitigations (KPTI) made the syscall about 3x slower, so I
guess it could be now comparable or even larger than the HPET reading.

-- 
Miroslav Lichvar