StarLink as source for NTP
-
@Mission-Ghost said in StarLink as source for NTP:
100.117.x.x.
You really do not need to worry about sharing addresses that are behind CGNAT.. Nobody can do anything with that info.
https://ipinfo.io/ips/100.117.0.0/16
My Verizon backup WAN is 100.124.148.197 right now.. If you traceroute it you will most likely get a Starlink user if you get anything at all..
But if you're worried.. just trying to save ya a little work.
-
@chpalmer Thanks; thinking it through this makes sense. This saves a lot of work. Redaction is a hassle.
-
Yup if you have checksum hardware offloading enabled it's normal to see checksum errors in a pcap.
-
You can upload a pcap here and I'll check it: https://nc.netgate.com/nextcloud/s/GFEbm2KqHTGQYjp
-
@stephenw10 said in StarLink as source for NTP:
You can upload a pcap here and I'll check it: https://nc.netgate.com/nextcloud/s/GFEbm2KqHTGQYjp
Done. Thank you!
-
Ok I see that. So that is filtered for ntp and after restarting ntpd?
-
That looks completely as expected. I would expect ntpd to use that fine if it's actually seeing the replies. Which it should!
I assume you're not seeing any of that blocked on the WAN?
-
@Mission-Ghost said in StarLink as source for NTP:
Ok, I have a pcap 'full' file. I ran the logs for about 7 minutes, which was long enough to get 377 reaches on my nist time server pool servers.
What is the full status shown on Status / NTP?
Also, do you have peer messages enabled in Services / NTP / Settings? Assuming so, did you check the system log for NTP messages?
-
It's filtered for 192.168.100.1 on the Starlink WAN interface port 123.
-
3+ hours since NTP restart:

Peer messages: yes

System Log>NTP messages from today:


No messages for these two IPs in System>General
I wonder why the pfBlocker VIP 10.10.10.1 is getting involved. Could this be a factor?
So I stopped pfBlocker and these records are now in the NTP log:

and RefID went to .XFAC. in the NTP status:

-
@stephenw10 said in StarLink as source for NTP:
I assume you're not seeing any of that blocked on the WAN?
I have no WAN rules, but also no logs in System>Firewall. I can set something up if you like.
-
Below is Grok’s assessment. Perhaps it has SpaceX specialized information? Are its assertions that:
pfSense uses an older ntpd implementation that doesn't fully associate or poll the dish's NTP server properly, even though one-off queries succeed.
And
This is a known compatibility quirk with pfSense's ntpd and the dish's NTP implementation. The daemon doesn't shift the reach register properly for ongoing polling.
pfSense time may still drift-correct slowly using the responses, but status shows unreachableAccurate?
The whole output:
The Starlink dish provides a Stratum 1 GPS-based NTP server at 192.168.100.1 when the Starlink router is in bypass mode (common with third-party routers like pfSense). Your packet capture confirms bidirectional NTP traffic (UDP port 123), so connectivity and basic responses work.
The "reach 0" issue in pfSense's NTP status (visible via
ntpq -por Status > NTP) occurs because pfSense uses an older ntpd implementation that doesn't fully associate or poll the dish's NTP server properly, even though one-off queries succeed.Quick Troubleshooting Steps
-
From the pfSense shell (Diagnostics > Command Prompt or SSH), run:
ntpdate -qu 192.168.100.1
This should return a valid stratum 1 response with low offset/delay, confirming the server works for basic queries. -
Ensure access to 192.168.100.1:
Most Starlink + pfSense setups require:- A Virtual IP (e.g., 192.168.100.2/24) on the WAN interface (Firewall > Virtual IPs).
- An Outbound NAT rule (Firewall > NAT > Outbound, set to Manual/Hybrid) sourcing from your LAN (or WAN) and translating to the virtual IP for destination 192.168.100.0/24.
- On the WAN interface: Reject DHCP leases from 192.168.100.1 (Interfaces > WAN > DHCP client configuration).
This makes the dish see queries as coming from the same subnet.
-
In Services > NTP:
Add 192.168.100.1 as a server (mark as Preferred if desired).
Restart the NTP service after changes.
Why Reach Remains 0 (and Workarounds)
If reach stays 0 despite traffic:
- This is a known compatibility quirk with pfSense's ntpd and the dish's NTP implementation. The daemon doesn't shift the reach register properly for ongoing polling.
- pfSense time may still drift-correct slowly using the responses, but status shows unreachable.
Recommended fixes:
- Keep 192.168.100.1 as primary/preferred, but add reliable public servers (e.g., time.google.com, pool.ntp.org) as fallbacks. pfSense will prioritize the local one when possible.
- The dish's low-latency GPS time is excellent for reducing jitter vs. internet NTP (especially with Starlink's variable latency).
- If you need perfect status reporting, consider a local NTP appliance or fallback to public pools only.
This setup works for many users accessing the dish for pings, stats, and NTP—though the built-in ntpd status often misreports reachability.
-
-
@elvisimprsntr said in StarLink as source for NTP:
I have it disabled and GPS+PPS disciplined NTP server is accurate on a Protectli FW4CI was not suggesting NTPd does not work. Rather chrony just does it better
- https://itsfoss.gitlab.io/post/chrony--an-alternative-ntp-client-and-server-for-unix-like-systems/
- Performance section of https://chrony-project.org/comparison.html
Which is why I use it rather than NTPd
@dennypage said in StarLink as source for NTP:
it's risk vs. reward, and each corporate entity, Netgate included, has to make their own decision based on their own circumstance.
@luckman212 said in StarLink as source for NTP:
I raised an issue on their GitLab, hope that isn't too annoying.
Which suggest
chrony is intended to be GPLv2 only and therefore not able to be combined and redistributed with Apache 2.0-licensed code.
Looks like supporting chrony via a pfsense optional package maybe the easiest legally. Which would work well for me.
-
@Patch said in StarLink as source for NTP:
I was not suggesting NTPd does not work. Rather chrony just does it better
No question about it. But I’m not sure it’s enough of a difference to justify it for Netgate.
Maybe I should write a package for it…

-
Chrony does not support GPS+PPS by itself. You also need GPSD to ingest the NMEA messages and PPS via SHM/SOCK to Chrony. There is also a known issue with GPSD reading PPS via the CTS pin vs the DCD pin, in spite of the GPSD source code claiming it can read PPS via CTS. Let's just say the maintainer of GPSD is not very accepting of constructive criticism or anyone suggesting otherwise.
You won't find those of us with a GPS+PPS stratum 0 source connected to directly to pfSense using Chrony. With the FreeBSD loader.conf setting (dev.uart.0.pps_mode="1"), NTPD can read PPS vis the CTS pin, which is populated on RJ45 serial COM ports. I get less than 2 usec offset/jitter.
IMHO, anyone who cares about accuracy, fault tolerance, security, and being a good citizen by not allowing individual clients to use public NTP servers, should be running their own local stratum 1 NTP server connected directly to a disciplined stratum 0 source. Public sources should only be used as a referee and/or fallback.
https://github.com/elvisimprsntr/pfsense-ntp-gps
ntpq -pn remote refid st t when poll reach delay offset jitter ============================================================================== o127.127.20.0 .GPS. 0 l 1 8 377 0.000 +0.000 0.002 time.nist.gov .POOL. 16 p - 64 0 0.000 +0.000 0.000 +192.5.41.40 .PHC. 1 u 11 64 377 41.558 +0.646 0.299 +192.5.41.41 .PHC. 1 u 48 64 377 41.491 +0.739 1.018 +132.163.96.1 .NIST. 1 u 58 64 377 50.814 +0.353 0.309Overview
On Linux, Chrony + GPSD is a common pairing for GPS‑based time sources. But on FreeBSD, this stack is not only unnecessary — it is objectively inferior for anyone requiring microsecond‑level precision, forensic defensibility, or stable PPS discipline.FreeBSD provides something Linux does not:
Native kernel‑level PPS discipline directly from UART pins.
No GPSD.
No SHM.
No SOCK.
No user‑space jitter.This document explains why Chrony + GPSD is the wrong architecture on FreeBSD, and why ntpd + kernel PPS is the correct, proven, and measurably superior solution.
-
Chrony Cannot Read PPS from Serial Hardware
Chrony has no ability to read PPS directly from:
UART CTS
UART DCD
GPIO
Any hardware pin
Chrony requires PPS to be delivered via:
SHM (shared memory) from GPSD
SOCK (chrony SOCK refclock) from GPSD
This means:
Chrony is always one layer removed from the PPS signal.
Every layer adds jitter. -
GPSD Is a Known Weak Link for PPS
GPSD claims to support PPS on both:
DCD
CTS
But in practice:
PPS Pin GPSD Behavior Notes
DCD Works reliably Most GPS modules expose PPS here
CTS Often fails or mis‑timestamps Despite GPSD claiming support
The GPSD maintainer has repeatedly rejected:
Bug reports
Reproducible test cases
Patches
Field evidence
This is why experienced timing engineers avoid GPSD unless absolutely required. -
FreeBSD Has Native PPS Support — No GPSD Needed
FreeBSD’s UART driver can timestamp PPS edges in the kernel, with nanosecond‑level precision.
Enabling PPS on CTS is a single loader.conf line:
dev.uart.0.pps_mode="1"
This gives you:
/dev/pps0 exposed directly by the kernel
PPS timestamps taken in interrupt context
No user‑space latency
No SHM jitter
No GPSD dependency
Direct kernel PLL discipline
This is the same architecture used by:
Telecom timing labs
Stratum‑0/1 reference clocks
Scientific instrumentation
GNSS timing receivers
4. NTPD + Kernel PPS Discipline Is Superior on FreeBSD
When ntpd uses kernel PPS, you get:
Hard PPS discipline (RFC 1589)
PPS edges discipline the kernel clock directly
NMEA sentences provide coarse time
PPS provides fine time
Kernel PLL keeps long‑term stability
Measured performance on real FreeBSD hardware
Your own system (J3710, FreeBSD 16‑CURRENT):
Offset: 2 µs
Jitter: 2 µs
Stability: rock‑solid
No GPSD, no SHM, no SOCK
This is exactly what a properly configured Stratum‑1 should deliver.
Chrony + GPSD cannot match this on FreeBSD.
5. Architecture Comparison
Diagram: Chrony + GPSD (Linux‑style stack)GPS Module
┌──────────────┐
│ NMEA + PPS │
└──────┬────────┘
│
▼
┌──────────────┐
│ GPSD │ ← timestamps PPS in user space
└──────┬────────┘
│ SHM/SOCK
▼
┌──────────────┐
│ Chrony │ ← disciplines clock using user-space PPS
└──────────────┘
Weaknesses:
PPS timestamped in user space
SHM jitter
GPSD reliability issues
No kernel PLL discipline
No direct hardware PPS path
Diagram: FreeBSD Native PPS + ntpdGPS Module
┌──────────────┐
│ NMEA + PPS │
└──────┬────────┘
│
▼
┌──────────────┐
│ FreeBSD UART │ ← timestamps PPS in kernel
│ pps_mode=1 │
└──────┬────────┘
│ /dev/pps0
▼
┌──────────────┐
│ Kernel PLL │ ← hard PPS discipline
└──────┬────────┘
│
▼
┌──────────────┐
│ ntpd │ ← uses kernel PPS + NMEA
└──────────────┘
Strengths:
PPS timestamped in kernel
Zero user‑space jitter
RFC‑1589 hard PPS discipline
No GPSD required
No SHM/SOCK
Microsecond‑level stability
6. Timing Comparison
Chrony + GPSD (FreeBSD)
Typical real‑world results:
Metric Typical Value
Offset 10–50 µs
Jitter 10–100 µs
Stability Variable
PPS Path User space
FreeBSD ntpd + Kernel PPS
Your actual measured results:
Metric Your System
Offset: 2 usec
Jitter: 2 usec
Stability Excellent
PPS Path Kernel
This is an order‑of‑magnitude improvement.
7. The Bottom Line
If you are running FreeBSD or pfSense and you have a real GPS+PPS source:
Do NOT use Chrony + GPSD.
It adds jitter, complexity, and failure modes. -
-
@elvisimprsntr said in StarLink as source for NTP:
Chrony does not support GPS+PPS by itself.Oops
Accidentally got side tracked on a tangent.- I use NTP servers to get network time.
- My understanding was the Starlink receiver provides a GPS based NTP server not a serial port pulse per second reference signal.
For these scenarios, my understanding is Chrony is simply better.
The rabbit hole I was not intending to go down is how best to building a Pulse per second base NTP server,
- Chrony can be configured to support PPS but software time stamping introduces significant jitter and offset.
- @elvisimprsntr the configuration you propose probably does reduce this jitter and offset on FreeBSB systems (but I have made no attempt to verify that)
- The recommended way to really reduce the offset and jitter is use a software defined pin on the on the NIC which is receiving requests and sending responses to NTP clients. This allows the PPS signal to be timestamped in hardware. Doing so eliminates the software as well as the hardware PCIe and interrupt delay jitter and offsets.
Getting back on topic. Chrony running on pfsense would significantly increase it's accuracy as a system time reference, but clearly not to the level of dedicated NTP server hardware.
-
@Patch said in StarLink as source for NTP:
My understanding was the Starlink receiver provides a GPS based NTP server not a serial port pulse per second reference signal.
I never implied Starlink provided a PPS signal.
@Patch said in StarLink as source for NTP:
Chrony can be configured to support PPS but software time stamping introduces significant jitter and offset.
Yes it can be configured to read PPS, but then you have to rely on a public NTP sever for a source of time data since Chrony by itself cannot digest NMEA messages. So if your WAN goes down, you have no way to maintain accurate time.
-
Hmm, the logs seem to agree that it's just not seeing the replies for some reason....

-
@Ramosel Do you also have pfBlockerNG package installed?
-
@tinfoilmatt said in StarLink as source for NTP:
Do you also have pfBlockerNG package installed?
Yes. Saw no issues in unified or alerts. Also checked Snort.
Hello! It looks like you're interested in this conversation, but you don't have an account yet.
Getting fed up of having to scroll through the same posts each visit? When you register for an account, you'll always come back to exactly where you were before, and choose to be notified of new replies (either via email, or push notification). You'll also be able to save bookmarks and upvote posts to show your appreciation to other community members.
With your input, this post could be even better 💗
Register LoginPrivacy Policy · Cookie Policy