Scalable PTPv2.1 two-step unicast server implementation

Default arguments are good for most of the cases. However, there is a lot of room for customisation:

/usr/local/bin/ptp4u -iface eth1 -workers 100 -minsubinterval 1us -monitoringport 1234

This will run ptp4u on eth1 with 100 workers and allowing 1us subscriptions. Instance can be monitored on port 1234

By default ptp4u runs http server serving json monitoring data. Ex:

$ curl localhost:8888 | jq
{
  "rx.delay_req": 0,
  "rx.signaling.announce": 0,
  "rx.signaling.delay_resp": 0,
  "rx.signaling.sync": 0,
  "subscriptions.announce": 1,
  "subscriptions.sync": 1,
  "tx.announce": 1,
  "tx.delay_resp": 0,
  "tx.follow_up": 1,
  "tx.signaling.announce": 0,
  "tx.signaling.delay_resp": 0,
  "tx.signaling.sync": 0,
  "tx.sync": 1,
  "utcoffset": 37,
  "worker.0.txtsattempts": 1
}

This returns manu useful metrics such as number of active subscriptions, tx/rx stats etc.

We were able to generate and consistently support over 1M clients with synchronization frequency of 1Hz.

We used a following setup:

• HPE DL380 G10 single CPU system.
• Arista 7010 network switch with 2 VLANs.
• Calnex Sentinel to monitor the precision over the PTP compared to the GPS source.
• Spirent N4U to generate test clients and verify timing correctness.

Here one can see 1000513 clients registered on ptp4u side, where:

• 512 are constantly verified for the timing violations.
• 1 is a Calnex Sentinel to verify precision effect from the load. The vertical line represents the beginning of the test.
• 1000000 generated using traffic generator feature. Spirent N4U showing 512 monitoring clients are working as expected: