Time synchronization

[lars_the_bear]

Hi folks

This is only tangentially a Gentoo question.

I just switched the Gentoo systems I recently installed, from using chrony to sntpd. I think chrony is just too much time-sync for a minimal system. The RHEL/Fedora systems I use all get chrony by default. Just out of curiosity, I looked at my Ubuntu installations, and saw they had systemd_timesyncd instead, which is something I had never even noticed before.

The source for Gentoo's version of sntpd say that:

"sntpd is a fork of Larry Doolittle's ntpclient with added daemon, syslog, and IPv6 support"

So it has some history, and there's a bit of documentation. systemd_timesyncd appears to be completely undocumented, with no indication where it came from.

Since sntpd and systemd_timesyncd do essentially the same thing and, so far as I can tell, in the same way, does anybody know where the are different? The systemd versions seems to do stuff with DBus, but I don't really understand what.

Does anybody know why, on a systemd system, I should prefer one implementation to the other? And, just out of idle curiosity, why the systemd folks wanted their own time sync daemon, when there are plenty of others around?

BR, Lars.

[pingtoo]

I think you answered your own question. may be because systemd_timesyncd talk over dbus. as far as I know no other time sync daemon talk over dbus.

out of topic, I am under impression you prefer no systemd from the udev alternative thread, so why now you consider systemd_timesyncd?

Personally I use openntpd, which is simple time sync daemon from bsd folks.

[lars_the_bear]

[szatox]

All ntp daemons attempt to synchronize the local clock with the network.
Which is an impossible task, so they're doing their best to approximate the ideal. That's the keyword: approximate. The correct solution does not exist, so we just do _something_ and call it good enough.

And this is where different implementations differ. They track different things and fail in different ways.
E.g. ntpd follows mean time from multiple sources and crashes if the drift becomes too big for it to compensate (but keeps working even without authoritative source), chrony follows the time from whtever source it considers to be the best but adjust times faster and can recover from failures, while systemd-timesyncd exists because systemd-folks must put their finger into every single dish
Never used sntpd

Anyway, in vast majority of cases any npt daemon is good enough, but they all cut some corners somewhere (out of necessity), so in your particular case one might be more reliable than the other, depending on the environment and your priorities.
_________________
Make Computing Fun Again

[pa4wdh]

There are quite some NTP implementations which differ quite a lot.

Main differences:
1) How they sync and cope with latency/failures
2) Support for hardware clocks to sync to

2 is of course only relevant if you want to build a time server using a reference clock (for example GPS), 1 depends on how accurate you need to be. From what i found in the documentation sntpd and systemd-timesync both just query a server and set the local clock to whatever the server is telling them, that might be good enough for you. (x)ntpd and chrony do their best to do a better job by keeping track of latency/jitter/drift and compensate for that.

There is a comparison page in the chrony project: https://chrony-project.org/comparison.html
I expect it to be biased towards "chrony is better", i also spotted some misinformation about ntpd (which i expect they didn't do on purpose), but i think it still might help you to make a choice.
_________________
The gentoo way of bringing peace to the world:
USE="-war" emerge --newuse @world

My shared code repository: https://code.pa4wdh.nl.eu.org
Music, Free as in Freedom: https://www.jamendo.com

[logrusx]

[lars_the_bear]

[pingtoo]

[logrusx]

[pjp]

[lars_the_bear]

[Zucca]

Chrony is what you would want if you want all the features you could possibly need... or almost.
I've gone trough testing some ntp client-daemons.
I think the most simple one is the one that OpenBSD folks keep up to date. Clockspeed is one that is also interesting.

Relevant topic.
_________________
..: Zucca :..

[Goverp]

FWIW I was using net-misc/openntpd until last week, when I gave up on it and switched to straight net-misc/ntp. The reason for changing was that at boot time openntpd timed out connecting to my local ntp daemon (running on a Raspberry Pi, itself using ntp to set its clock from the uk.pool.ntp.org). Result of this was that for some reason the clocks had drifted way out of sync (more than 10 seconds). openntpd then starts synchonizing in very small steps every so often, and as I don't use my machines long enough in one sitting, they never caught up. Now with normal ntp configured with

[lars_the_bear]

[Zucca]

[Koyan]

[Koyan]

[Goverp]