Chrony, PTP, Network Time Security (NTS, NTPsec) to replace unsecure/old NTP (ntpd)

Sergei_Shablovsky

@johnpoz said in Network Time Security (NTS, NTPsec) to replace unsecure/old NTP (ntpd):

@sergei_shablovsky said in Network Time Security (NTS, NTPsec) to replace unsecure/old NTP (ntpd):

this mean double PSU, double Eth connection, best available servers hardware, memory, enterprise (mean big MTBF hours), etc...

If your worried about NTP in the enterprise.. I highly doubt you would be doing it on pfsense to be honest.. More than likely you would have some ntp appliance or multiple ones most likely, etc.

Sorry but if your goal is ntp.. In an enterprise I sure wouldn't be running it on my firewall/router ;) And more likely than not I wouldn't be setting up any hardware - what would be done is get a box that is designed to provide NTP to the enterprise..

I clearly understand Your point.

But anyway, the so-called “NTP Server” (in case that they cannot obtain PPS thru the radio waves, but only thru the GPS receiver) - is no more than the same 'nix system for embedded platforms, that running inside the device.
This is exactly the same as having separate x86_64 server, but less flexible and more dependent on a yearly payment for device developer for firmware upgrade.

Am I wrong here?

DaddyGo

@sergei_shablovsky said in Network Time Security (NTS, NTPsec) to replace unsecure/old NTP (ntpd):

Sorry my misstyping, I mean that’s phrase made by myself. :)

Furthermore...

Everything you need to know about NTP at enterprise and industrial level can be found here:

https://www.meinbergglobal.com/

Doing NTP well is not easy, because, say, one temperature dependency of crystal can throw the whole thing in the trash.
(Not to mention the delay of the NTP distribution network)

That's why this hardware costs so damn much, stability - stability - compensation and stability again.
on pfSense is not worth thinking about it...
(if you want to have close to exact time on your network, choose something like this:
https://nguvu.org/pfsense/network%20time%20protocol%20(ntp)/ntp-server/)

More for, say, data centre switches or audio systems, bank App, stock exchange, credit card schemes, NASA :)) - ......it's a big question really...

or PTP (AES67, DANTE, digital audio word clock, etc:
https://en.wikipedia.org/wiki/Precision_Time_Protocol

Sergei_Shablovsky

@daddygo said in Network Time Security (NTS, NTPsec) to replace unsecure/old NTP (ntpd):

Doing NTP well is not easy, because, say, one temperature dependency of crystal can throw the whole thing in the trash.
(Not to mention the delay of the NTP distribution network)
That's why this hardware costs so damn much, stability - stability - compensation and stability again.
on pfSense is not worth thinking about it...
(if you want to have close to exact time on your network, choose something like this:
https://nguvu.org/pfsense/network%20time%20protocol%20(ntp)/ntp-server/)

Please read whole docs carefully.

Here we discuss scheme “pfSense driver receive PPS from local connected GPS device thru COM port”, and nothing about linking to the hardware (CPU frq generator, etc) on which exactly pfSense working.

You may connect pfSense to small Garmin marine GPS with 8-12 channels, or more complicated debice like listed several posts above,- anyway the results (PPS signal) come to COM port of pfSense server.
Let’s to note when You have the “Ethernet” port on Your stand-alone Time-Source device, this mean inside of this device are some firmware that realise ... the same NTP server. And in this case You have another one point of delay because need time for converting PPS signal to answers from NTP server inside the device.

This “stand-alone time sync devices” born a lot of years ago, and still live now only because certification system for health, financial, military industry exist. They are not much more than specialized computer device with an GPS and RF receivers modules.
The extremely accurate GPS/radio receiver and robust bullet-proof engineered construction and a bunch of output connectors, - there are only one advantage of this devices.

Cheers ;)

Sergei_Shablovsky

@daddygo said in Network Time Security (NTS, NTPsec) to replace unsecure/old NTP (ntpd):

@sergei_shablovsky said in Network Time Security (NTS, NTPsec) to replace unsecure/old NTP (ntpd):
Doing NTP well is not easy, because, say, one temperature dependency of crystal can throw the whole thing in the trash.
(Not to mention the delay of the NTP distribution network)

That's why this hardware costs so damn much, stability - stability - compensation and stability again.
on pfSense is not worth thinking about it...
(if you want to have close to exact time on your network, choose something like this:
https://nguvu.org/pfsense/network%20time%20protocol%20(ntp)/ntp-server/)

The device here are just for lab using or experiment: the GPS receiver are for hobbyist, and the computing module have no chances to compare with even old 10+ years IBM, Dell, HP servers ;)

DaddyGo

@sergei_shablovsky said in Network Time Security (NTS, NTPsec) to replace unsecure/old NTP (ntpd):

Please read whole docs carefully.

I don't have a problem with what you've written, but thanks the call for attention.

I have been working on timing theme (PTP stuffs) for years and I thought I would share with you the tools we use in our own radio station network.

https://www.meinbergglobal.com/english/products/ptp-ieee-1588.htm

https://dev.audinate.com/GA/dante-controller/userguide/webhelp/content/clock_synchronization.htm

Patch

I don’t doubt hardware optimised for time keeping will do better than hardware optimised for firewall functionality.

I don’t think that’s relevant though. The issue is chrony provides better functionality on whatever hardware it runs on. It’s simply better at it’s job, so given the choice, it is the preferred option.

But then again both of the above are answer to the the wrong question. A more relevant question is: is improving the time functionality of high enough priority to actually be done by a company who sells expertises in firewalls. I suspect the answer to this question is no, it will be upgraded when added upstream.

So unless someone outside of Netgate is willing and able to implement and test a chrony port, I can’t see it happening.

JKnott

@daddygo said in Network Time Security (NTS, NTPsec) to replace unsecure/old NTP (ntpd):

Doing NTP well is not easy, because, say, one temperature dependency of crystal can throw the whole thing in the trash.

If you're relying on a crystal, you're doing it wrong. NTP servers are supposed to be traceable back to something called International Atomic Time, which is the average of several atomic clocks around the world. The NTP software averages out the variations and if you have multiple sources (you should have at least 3), your time will actually be better than a single source. That said, however, hardware quality may affect jitter.

Sergei_Shablovsky

@patch said in Network Time Security (NTS, NTPsec) to replace unsecure/old NTP (ntpd):

I don’t doubt hardware optimised for time keeping will do better than hardware optimised for firewall functionality.

Totally agree. In previous replies I just note that most of the modern devices are just “PPS signal source (GPS, RF) + embedded SoC that realize NTP server”

I don’t think that’s relevant though. The issue is chrony provides better functionality on whatever hardware it runs on. It’s simply better at it’s job, so given the choice, it is the preferred option.

But then again both of the above are answer to the the wrong question. A more relevant question is: is improving the time functionality of high enough priority to actually be done by a company who sells expertises in firewalls. I suspect the answer to this question is no, it will be upgraded when added upstream.

Thank You for most relevant reply here ;)

So unless someone outside of Netgate is willing and able to implement and test a chrony port, I can’t see it happening.

May be I have time in this wintertime ;)

P.S. Chrony able to using NTS/NTPsec.

P.P.S.

From Netgate Official docs:

Troubleshooting Clock Issues

Time and clock issues are relatively common on hardware, but on firewalls they are critical, especially if the firewall is performing tasks involving validating certificates as part of a PKI infrastructure.

...

Not only will getting this all in line help with critical system tasks, but it also ensures that the log files on the firewall are properly timestamped, which aids with troubleshooting, record keeping, and general system management.

DaddyGo

@jknott said in Network Time Security (NTS, NTPsec) to replace unsecure/old NTP (ntpd):

If you're relying on a crystal, you're doing it wrong.

Yup,

I did not mean that I produce the time source itself with a crystal, say a VCO

unless I was referring to the fact that all computing devices follow some basic clock, e.g. CPU, BUS, RAM cycles, etc.

So there are a thousand points where time can be lost...

Of course, what you describe is the right approach, but it is also pointed out wherever time is involved, for example here:
(this is one of the behaviors of the NTPd and what would be good) + so this should not be news

++++edit:

BTW:
https://www.microsemi.com/product-directory/3425-timing-synchronization

HSO with Rubidium, OCXO, TCXO, Quartz = crystal :)
https://timetoolsltd.com/atomic-clocks/high-stability-oscillators/

JKnott

@daddygo

I have 5 sources, 3 stratum 1 and 2 stratum 2. One thing some people don't realize is the math that goes on to calculate the transit delay and then the error from the source. They're described in this article. I had to show that to a co-worker a couple of years ago. He thought each hop was delay from the one it connected to.

GPS is an excellent source as it traces back to IAT, through atomic clocks on the satellite. There's also WWVB. There were a couple of other methods that are pretty much gone now. One was the old 2G CDMA cell network, which used extremely precise time on the phones and the NTSC analog TV signal, where the colour burst frequency was tied to an atomic clock and some stations (PBS) provided the time of day in the vertical blanking interval. Even short wave radio broadcasts from WWV or CHU can be used, though their short term stability is not as good as WWVB. An NTP server that's not traceable back to IAT is supposed to be stratum 15.

BTW, here's a free book from the NIST about time.
From Sundials To Atomic Clocks

Does anybody really know what time it is?

JKnott

@daddygo said in Network Time Security (NTS, NTPsec) to replace unsecure/old NTP (ntpd):

HSO with Rubidium, OCXO, TCXO, Quartz = crystal :)

Yep, the crystal will be synced to the source and provide the correct time should you lose the connection to the source.

JKnott

@jknott

Forgot to mention, my background is in the telecom industry. Prior to IP becoming so popular, the phone network was based on time division multiplexing, which required precise synchronization. The way this was done was to include the timing in the signalling. At the company I worked for, LORAN C was used as the primary source. However, that provided a time base only and not time of day. Some of the people I worked with didn't understand the difference.

DaddyGo

@jknott said in Network Time Security (NTS, NTPsec) to replace unsecure/old NTP (ntpd):

Forgot to mention, my background is in the telecom industry.

In a way, I am also a telecom professional, I used to build telecommunication microwave networks and antenna systems
(I got a degree in RF engineering, huhu many many years ago :-))

Now I am the chief engineer of a URH-FM network of several radio stations.
(yeah, but we currently have several satellite broadcasting cars, where timing is also important)

BTW:
Thanks for the technical guide (NIST) I have not come across this before

bingo600

In a "Big Enterprise" where timestamping is critical , nobody would use the firewall as their NTP Master.

They would use dedicated hardware , as the Mentioned Meinberg above, or Ie.
Symmetricom.
https://prostudioconnection.com/products/symmetricom-syncserver-s300-gps-ntp-network-time-server-atomic-clock-receiver-refurbished-1

They would have 2..4 of these boxes , spread over multiple sites ,and all in a "Peer setup", they might even use a couple of "selected" external NTP servers, for extra reduncancy. But their primary trust would be their internal dedicated NTP servers.

You can even get some of the bxes doing both NTP & "SYSPLEX" ... @JKnott

Dealing with timestamping firewall loglines and/or Certificate validation would easily be fulfilled, by letting the firewall be a Client to the "Inside coproprate NTP setup".

Your normal PC/server Xtal is NOT "Time-nut" worthy , and it will drift quite a lot depending on environment temperature.
The Interrupt Latency would also be variable , especially with the CPU ability to scale frequency up & down , and will contribute to the Jitter.

A dedicated NTP server would often have 1-PPS timestamping in hardware , and if the buyer isn't going "Cheap" , there will be an OCXO or even a Rubidium clock source inside.
The need for the OCXO/Rubi is typically for improving the Holdover period (aka if the GPS signal is lost). A TCXO would be enough for keeping the "Specs" if one would trust that the GPS signal was always present.

PHK did a "Super precise NTP server implementation" using the Soekris
https://www.febo.com/time-freq/ntp/soekris/index.html

And the "Magic" was HW-timer 1-PPS timestamping , in the ELAN CPU used in the Soekris. Helped
Ohh his work on the BSD Kernel timestamping, didn't hurt either.
But that would be available for any BSD HW.

Edit:
This is the drift on my Atom270 based NTP Server, using a "simple" Xtal

$ cat /var/lib/ntp/ntp.drift 
26.975

/Bingo

Sergei_Shablovsky

@bingo600

Dear pfSense friends! (May is call You like that ? ;)

Personally I thankful to all here and each of You about suggestions and professional-grade level of discussion! A lot of interesting info for others and some - may be interesting for Pro-grade SysAdmins.

Only a few users here on forum have dedicated stand-alone Time-source device, like You refer to in several replies ago.
So, let me propose to change our direction back to the Chrony with NTS/NTPsec as replace of old/unsecured NTP. (In form of additional package with GUI for pfSense).

Cheers

DaddyGo

@sergei_shablovsky said in Network Time Security (NTS, NTPsec) to replace unsecure/old NTP (ntpd):

So, let me propose to change our direction back to the Chrony

Uhum

as someone wrote above this needs to be ported, I don't think it will be included otherwise, may need a persistent and dedicated programmer with a lot of time,....hihihi

You wrote you may have time in the winter, we would welcome such an initiative

(note: as a system administrator say, it's COVID time again, ergo more work, not to mention Ransomware + APTs

johnpoz

@daddygo said in Network Time Security (NTS, NTPsec) to replace unsecure/old NTP (ntpd):

as someone wrote above this needs to be ported

I think 3.5 is available since freebsd 12? Just in the normal freebsd packages.. So it should be possible without having to build or port, etc. Latest is 4.1?

DaddyGo

@johnpoz said in Network Time Security (NTS, NTPsec) to replace unsecure/old NTP (ntpd):

I think 3.5 is available since freebsd 12?

Ah, right it's already 4.1 currently...
https://www.freebsd.org/cgi/ports.cgi?query=Chrony&stype=all

++++edit:
Then all we need is PHP, Netgate and you're done :-)

johnpoz

@daddygo said in Network Time Security (NTS, NTPsec) to replace unsecure/old NTP (ntpd):

Then all we need is PHP, Netgate and you're done

Easy peasy ;) @Sergei_Shablovsky can you knock that out this afternoon ;) hehehe

Sergei_Shablovsky

@daddygo said in Network Time Security (NTS, NTPsec) to replace unsecure/old NTP (ntpd):

@johnpoz said in Network Time Security (NTS, NTPsec) to replace unsecure/old NTP (ntpd):

I think 3.5 is available since freebsd 12?

Ah, right it's already 4.1 currently...
https://www.freebsd.org/cgi/ports.cgi?query=Chrony&stype=all

++++edit:
Then all we need is PHP, Netgate and you're done :-)

As You may see both for ntpd (FreeBSD) and Chrony (FreeBSD, offsite) a lot of settings are common, so most users that need NTP service on pfSense would have short adoption time and not to be disappointed by a lot of new options in NTP GUI.

The main question is still “because of ntpd was outdated is replacement to Chrony in pfSense NTP service needed, or better to keep both ntpd and Chrony as additional pfSense package”.

Anyway time of ntpd gone away (like most of old protocols) and one day this question pop up again. Why need to be so sticky to old protocol? Chrony able to keep old-fashioned ntpd clients well without any issues, Chrony already exist in FreeBSD 12 and 13...