Warning
This is optional! Proceed with caution and at your own risk!
While there is quite some information available how to properly set up PTP on your Raspberry device, e.g.,
- https://github.com/tiagofreire-pt/rpi_uputronics_stratum1_chrony/blob/main/steps/advanced_system_tuning.md
- https://gpsd.gitlab.io/gpsd/gpsd-time-service-howto.html#_providing_local_ntp_service_using_ptp
- https://docs.fedoraproject.org/en-US/fedora/latest/system-administrators-guide/servers/Configuring_PTP_Using_ptp4l/#sec-Serving_NTP_Time_with_PTP
- https://sourceforge.net/p/linuxptp/mailman/linuxptp-devel/thread/1424738292.9759.53.camel%40intel.com/?page=1
not everything works right out of the box with my particular setup, a system with a highly relaibale clock, driven by GNSS with PPS, and a set of clients which should be serverd by PTP due to its higher precision of the time signal; thus, I've compiled this 'best of'. – Thank you very much to the authors of above documentation for their valuable insights, particularly @tiagofreire-pt and the contributors to the linuxptp-devel mailing list!
(Adjusted from https://github.com/tiagofreire-pt/rpi_uputronics_stratum1_chrony/blob/main/steps/advanced_system_tuning.md)
Every ethernet adaptor uses the coalescence method to gather packets and sent them on a bulk, for better thoughput efficiecy.
But this method introduces valuable latency and jitter on NTP or PTP-E2E packets.
Check the defaults:
sudo ethtool -c eth0
Coalesce parameters for eth0:
Adaptive RX: n/a TX: n/a
stats-block-usecs: n/a
sample-interval: n/a
pkt-rate-low: n/a
pkt-rate-high: n/a
rx-usecs: 49
rx-frames: n/a
rx-usecs-irq: n/a
rx-frames-irq: n/a
tx-usecs: 49
tx-frames: n/a
tx-usecs-irq: n/a
tx-frames-irq: n/a
rx-usecs-low: n/a
rx-frame-low: n/a
tx-usecs-low: n/a
tx-frame-low: n/a
rx-usecs-high: n/a
rx-frame-high: n/a
tx-usecs-high: n/a
tx-frame-high: n/a
CQE mode RX: n/a TX: n/a
Both rx-usecs and tx-usecs have a value of 49 microseconds.
We'll set both to the mininum accepted by the Raspberry Pi ethernet driver:
sudo ethtool -C eth0 tx-usecs 4
sudo ethtool -C eth0 rx-usecs 4
To revert to the default values:
sudo ethtool -C eth0 tx-usecs 49
sudo ethtool -C eth0 rx-usecs 49
This could shave circa 40 usecs of response time over the chrony ntpdata statistics. Huge improvement on a NTP setup that has hardware timestamping and its higher accuracy.
To make it persistent over restarts, simply create a systemd service for eth0_coalescence:
sudo nano /etc/systemd/system/eth0_coalescence.service
Add this:
[Unit]
Description=Setting Speed
Requires=network.target
After=network.target
[Service]
ExecStart=/usr/sbin/ethtool -C eth0 tx-usecs 4 rx-usecs 4
Type=oneshot
[Install]
WantedBy=multi-user.target
Then, enable and start the eth0_coalescence service:
sudo systemctl enable --now eth0_coalescence.service
(Adjusted from https://github.com/tiagofreire-pt/rpi_uputronics_stratum1_chrony/blob/main/steps/advanced_system_tuning.md)
Raspberry CM4 and Pi5 have a PTP hardware clock within their Ethernet chips, so we leverage those to have another high performance reference clock in ntpsec.
Here is our setup and the related time modes (TAI: International Atonic Time (without leap seconds), UTC: Coordinated Universal Time (incl. leap seconds):
GNSS --> gpsd --> chrony/ntpsec --> NTP --> phc2sys --> ptp4l --> PHC --> [network] --> PHC --> ptp4l --> phc2sys --> SHM --> ntpsec
time
type:TAI UTC UTC UTC UTC TAI TAI TAI TAI UTC UTC UTC
As ntpsec(or ntpfor that matter) does not synchronise the NIC clock PHC itself, we need to use phc2sys to sync the system time with PHC and leverage ptp4l to provide the right number of leap seconds (currently 37), thus taking care of the conversion UTC-TAI and vice versa. – Please ensure that you have a reasonaly current version of linuxptp, e.g., https://packages.debian.org/source/testing/linuxptp / https://salsa.debian.org/multimedia-team/linuxptp !
sudo apt update && sudo apt install linuxptp -y
On the server, which has a highly precise time from its GNSS and PPS connection via ntpsec, we create a new file /etc/linuxptp/ptp4l.conf and prepare it to provide the exact time available from the system clock to the PHC, but also make sure that the system clock is not adjusted back from the PTP side:
sudo nano /etc/linuxptp/ptp4l.conf
[global]
# only syslog every 1024 seconds
summary_interval 10
# use one of the following values for clock accuracy:
# 6 The clock node synchronizes its time to the master reference time source.
# PTP assigns a time table to the clock node. A clock node with time class 6
# cannot become a member clock of any other clocks in the domain.
# 7 The former time class is 6. The clock node cannot synchronize its time to
# a time source. It enters the reappointment mode and meets the reappointment
# conditions. PTP assigns a time table to the clock node. A clock node with
# time class 7 cannot become a member clock of any other clocks in the domain.
# 13 The clock node synchronizes its time to a time source. ARB assigns a time
# table to the clock node. A clock node with time class 13 cannot become a
# member clock of any other clocks in the domain.
# 14 The former time class is 13. The clock node cannot synchronize its time to
# a time source. It enters the reappointment mode and meets the reappointment
# conditions. ARB assigns a time table to the clock node. A clock node with
# time class 14 cannot become a member clock of any other clocks in the domain.
# 52 The clock node with time class 7 becomes optional clock A because it does
# not meet the reappointment conditions. A clock node with time class 52 cannot
# become a member clock of any other clocks in the domain.
# 58 The clock node with time class 14 becomes optional clock A because it does
# not meet the reappointment conditions. A clock node with time class 58 cannot
# become a member clock of any other clocks in the domain.
# 187 The clock node with time class 7 becomes optional clock B because it does not
# meet the reappointment conditions. A clock node with time class 187 can
# become a member clock of another clock in the domain.
# 193 The clock node with time class 14 becomes optional clock B because it does
# not meet the reappointment conditions. A clock node with time class 193 can
# become a member clock of another clock in the domain.
# 248 Default time class value.
# 255 Clock node operating in slave-only mode.
clockClass 6
# use one of the following values for clock accuracy:
# 0x20 25ns 0x24 2.5us 0x28 250us 0x2c 25ms 0x30 10s
# 0x21 100ns 0x25 10us 0x29 1ms 0x2d 100ms 0x31 more than 10s
# 0x22 250ns 0x26 25us 0x2a 2.5ms 0x2e 250ms 0xfe unknown
# 0x23 1us 0x27 100us 0x2b 10ms 0x2f 1s
clockAccuracy 0x22
clock_servo linreg
free_running 1
# use one of the following for time source:
# 0x10 atomic clock 0x30 terrestrial radio 0x50 NTP 0x90 other
# 0x20 GPS 0x40 PTP 0x60 hand set 0xa0 int. oscillator
timeSource 0x20
[eth0]
#delay_mechanism Auto
network_transport L2
Now, we simply follow https://salsa.debian.org/multimedia-team/linuxptp/-/blob/master/debian/README.Debian?ref_type=heads :
Create a systemd service for ptp4l on intercafe eth0:
sudo systemctl enable ptp4l@eth0 sudo systemctl start ptp4l@eth0
Leave the systemd service for ptp4l@eth0 as is.
Create a systemd service for phc2sys:
sudo systemctl enable phc2sys@eth0
Adjust the systemd service file for phc2sys to adjust the PTP hardware clock, based on the system clock:
sudo nano /lib/systemd/system/[email protected]
[Unit]
Description=Synchronize system clock or PTP hardware clock (PHC)
Documentation=man:phc2sys
After=ntpd.service
[email protected]
[email protected]
Before=time-sync.target
[Service]
Type=simple
ExecStart=/usr/sbin/phc2sys -a -rr
[Install]
WantedBy=multi-user.target
Then, enable and start the phc2sys service for interface eth0:
sudo systemctl start phc2sys@eth0
One might think that [email protected] could also leverage the NTPSHM service (i.e., ExecStart=/usr/sbin/phc2sys -a -rr -E ntpshm -M 2 like on the clients' side, below), but this would leave the PHC unsynchronized and create an offset by those nasty 37 seconds of time difference between TAI and UTC between the clocks, as NTPSHM is only propagating the current time to the SHM2 memory slot and not synchronizing the PHC incl. the UTC-to-TAI adjsutment.
On the client, create a new file /etc/linuxptp/ptp4l.conf just with just this:
sudo nano /etc/linuxptp/ptp4l.conf
[global]
summary_interval 10 # syslog only every 1024 seconds
clock_servo linreg # linear regression (do _not_ use ntpshm here!)
[eth0]
delay_mechanism Auto
Again, follow https://salsa.debian.org/multimedia-team/linuxptp/-/blob/master/debian/README.Debian?ref_type=heads :
Create a systemd service for ptp4l:
sudo systemctl enable ptp4l@eth0 sudo systemctl start ptp4l@eth0
Leave the systemd service for ptp4l@eth0 as is.
Create a systemd service for phc2sys:
sudo systemctl enable phc2sys@eth0
Adjust the systemd service for phc2sys:
sudo nano /lib/systemd/system/[email protected]
[Unit]
Description=Synchronize system clock or PTP hardware clock (PHC)
Documentation=man:phc2sys
After=ntpd.service
Requires=ptp4l.service
After=ptp4l.service
#Before=time-sync.target
[Service]
Type=simple
ExecStart=/usr/sbin/phc2sys -a -rr -E ntpshm -M 2
[Install]
WantedBy=multi-user.target
The configuration of phc2sys is not quite straightforward here: with the option -E ntpshm -M 2, it does not change the systemclock directly but provides the time from PHC via phc2sys with correct leap secands thanks to ptp4l via SHM2 for chrony/ntpd/ntpsec to mix&match. – Do not do this in ptp4l.conf!
Then, enable and start the phc2sys service:
sudo systemctl start phc2sys@eth0
First check with sudo ntpshmmon if SHM2 is properly propagated.
Now, add this new refclock 'ntpsec' configuration with
sudo nano /etc/ntpsec/ntp.conf
refclock shm unit 2 refid PTP
Now check with ntpmon that your new refclock is working properly.