[time-nuts] Precision Time Protocol – Windows 10 implementation

[Tim Lister]

On Thu, Aug 8, 2019 at 8:07 AM Adam Kumiszcza <akumiszcza at gmail.com> wrote:
>
> Hi everybody! My first post here, I hope the subject is adequate for this
> mailing list.
>
> I'm using a tiny layer 1 NTP server consisting of Raspberry Pi 3B+ with
> Ublox MAX-M8Q expansion board providing GNSS (currently GPS, Galileo and
> Glonass, sometimes I switch to Beidou, too) reference with PPS + a simple
> patch antenna hanging near the window. Offset, jitter and rms are most
> often smaller than 1 µs. The server is included in NTP pool.
>
> I'm using several Windows 10 machines on the same LAN, all using NTP client
> software from Meinberg. The typical offset and jitter in those are about
> 100-500 µs. I would like to make it lower. I've heard that one of the
> latest versions of Windows 10 provides support for PTP protocol (IEEE
> 1588v2). Hence my questions:
>
> 1. Did anybody try using it on Windows 10 already? From what I can tell
> right now, there's only a demonstration web site showing it works when
> using Ubuntu Linux as a virtual machine [1]. Or should I wait a bit for
> more robust implementation?

EndRun Technologies
(https://endruntechnologies.com/products/grandmaster-clocks/ptp-slaves)
make mention of a Windows PTP slave - not sure if it's the same
version you have found. It does also make mention of the fact that
precision is limited on Windows depending on version although from
what I gather from the satsignal.eu that precious on Windows improved
with 7 or later.

> 2. Do I guess correctly that current implementation of PTP works only on
> Ethernet? Some of the computers are on wlan.

For deterministic timing I think it has to be Ethernet and as outlined
in their white paper (https://endruntechnologies.com/pdf/PTP-1588.pdf)
it works best if you have hardware timestamping in the NICs at each
end and IEEE1588 compatible network switches (and even when the claim
to support it, apparently they don't always do it properly:
https://www.linkedin.com/pulse/ptp-support-high-end-ethernet-switches-heiko-gerstung).
You may be able to do some Pi-to-Windows PTP experiments with a
crossover cable. Not sure if the Pi has the hardware timestamping,
seems unlikely given the low cost of the Pi's as in the 3B+ (and below
I believe) the Ethernet is provided by a combined Ethernet/USB2 hub
and limited by the speed of the USB2.

> 3. Is there any other way of making the clients' time offset lower?

NTP works best if the network delay is symmetric and stable - this
would also rule against wlan I'm afraid.

>
> I've already tried setting PTP on the Raspberry Pi 3B+. It needed a kernel
> recompilation with a patch I found here: [2], but I'm not sure it works
> correctly.
>

>From what I gather the Linux implementation can be either a slave or
act as the "grandmaster" without having to buy an expensive external
grandmaster clock (the ones I've seen from e.g. Microsemi, Brandywine,
Meinberg etc are of the "Get a quote" variety which tends to mean
"expensive")

>
> Thanks in advance for any suggestions!
>
> Adam Kumiszcza
>

Cheers,
Tim