NTP Performance

Gary E. Miller gem at rellim.com
Sun Nov 24 22:17:08 UTC 2019


On Sun, 24 Nov 2019 07:37:13 +0100
ASSI via devel <devel at ntpsec.org> wrote:

> Gary E. Miller via devel writes:
> > There is a gpsd program in the contrib/ directory.  It tests your
> > CPU granularity.  On a Raspberry Pi that is about 52 ns.  Worse
> > on an Intel chip.  
> The actual granularity on RasPi can't be better than 52ns (the clock
> it's based on is 19.2MHz) and you can determine it rather precisely
> since you can read quite a bit faster than the clock actually ticks.

Yes, exactly what clock_test does.  And clock_test does more!

Also check out attic/clocks in NTPsec.

> On Intel it is typically around 17ns if you use TSC, but the clock
> it's based on is 200MHz, so you read slower than the clock ticks.  If
> you use HPET as the clock source (14.318MHz typically, although only
> specified as
> >=10MHz), you'll see around 72ns granularity.  

Or just use clock_test and see.  Theory is great, but I trust data.

The granularity is important, but the fun part is the spread of results.
An interrupt here, a CPU swap there, and it gets real bad real fast.

ntpd could do a better job of tossing out wild measurements.  Maybe someday.

Gary E. Miller Rellim 109 NW Wilmington Ave., Suite E, Bend, OR 97703
	gem at rellim.com  Tel:+1 541 382 8588

	    Veritas liberabit vos. -- Quid est veritas?
    "If you can't measure it, you can't improve it." - Lord Kelvin
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 851 bytes
Desc: OpenPGP digital signature
URL: <https://lists.ntpsec.org/pipermail/devel/attachments/20191124/20506864/attachment.bin>

More information about the devel mailing list