pfSense NTP Fluctuating Offset

[argonaut]

Hi

My first post here as I am a pfSense noob and have just installed on what I thought is an overpowered system, but I am getting poor NTP performance, at least in comparison with what I am used to…

Hardware
Gigabyte H370 WiFi Mini-Itx MB with (2) Intel integrated NIC’s :-
i219-v as WAN
i211 as LAN
Intel Celeron G4920 CPU @ 3.20GHz
120GB M.2 SSD
4 GB RAM

Software
pfSense version 2.4.4-Release-p2
FreeBSD 11.2-Release-p6
No Packages installed

NTP Setup:-

pfSense ntp configuration file

tinker panic 0
Orphan mode stratum
tos orphan 12

Upstream Servers
server 10.4.0.14 iburst maxpoll 9 # a GPS time server on the LAN
server 10.4.0.16 iburst maxpoll 9 # a PPS time server on the LAN
server 10.4.0.17 iburst maxpoll 9 # a PPS time server on the LAN
server 10.4.0.7 iburst maxpoll 9 # a Windows PC always on, local LAN
server 10.4.0.18 iburst maxpoll 9 # a Synology Disk Station on 16 hrs/day

statsdir /var/log/ntp
logconfig =syncall +clockall
driftfile /var/db/ntpd.drift
restrict default kod limited nomodify nopeer notrap
restrict -6 default kod limited nomodify nopeer notrap
interface ignore all
interface ignore wildcard
interface listen igb0

Setup Information

The offset fluctuates wildly swinging from +0.100ms to -0.750ms in the space of one hour.
Offset is much larger than on other nodes on the network. By way of comparison:-
ntpd 4.2.8p12 running on Windows has a stable offset of around 0.011ms (11us)
ntpd 4.2.8p11 running on a Synology Diskstation DS713+ using an Intel Atom D2700 has a stable offset of around 0.009ms

Is this a hardware issue? e.g. CPU throttling or NIC handoff?
Is it a software configuration issue? Polling interval too long or Firewall configuration or Driver support?
Is it a BIOS configuration issue?

I have read that temperature can affect the offset – how sensitive to temperature is NTPD?

How does the file structure work?
There is no file at /etc/ntp as is normal with Linux?
If you edit the file at /var/etc/ntpd.conf it has no effect?

Sorry for all the noob questions but I think I must be doing something fundamentally wrong somewhere...

Argonaut

[argonaut]

I am still experiencing this issue and have not been able to find any satisfactory explanation or resolution to the problem.

As you can see from this graph:-

The offset definitely varies with room temperature and there is not that much variation really. 25 deg Celsius when the Air Conditioner gets turned on of a morning and 27.5 degrees Celsius when it is turned off at night.

In fact the offset tracks room temperature pretty accurately.

Now this is not a surprise as that's what I've read.

But what has got me baffled is why NTPD can't maintain the offset in a sub 100 micro second range as can other computers and devices on the network?

Any help greatly appreciated :)

PS some more info if this helps a diagnosis:-

[stephenw10]

For reference I see similar results on everything I have logging that.

Perhaps it's a display issue rather than an actual fault. Or other systems are reporting better numbers than they are actually seeing. Though that seems unlikely across such a range of systems.

You actually need better than 1ms accuracy?

Steve

[argonaut]

Thanks Steve for your reply.

I guess the accuracy thing is relative. For some people 1ms is great but when you have invested in some Stratum 1 NTP network servers which have a delay of around 100 us (sorry don't have a micro second symbol on my keyboard), then I expect NTPD to do better.

I don't understand why the implementation of ntpd in FreeBSD/pfSense is not as accurate as the implementation I have in Linux or even on my Windows machines.

I fired up a second machine with pfSense on it which has a serial port on the MB. I connected a Garmin 18x LVC with PPS and (somewhat) unsurprisingly the offset operates in a much narrower range (see below)

[stephenw10]

Mmm, I agree it seems odd. Have you been able to test it in FreeBSD directly?

If it is something we are doing in pfSense we could dig into it but if it's something FreeBSD does it would need to be reported upstream really.

Steve