timing goes crazy with Ethernet traffic
Daniel Florin has reported the issue below. Earlier, a very similar problem was reported by Peter Jansweijer (issue 926) on his Kintex7 WR nodes. However, at that time we were not able to reproduce it in our test system.
-- to: white-rabbit-dev@ohwr.org
-- date: 15 January 2015 at 17:42
We are currently experimenting with two WR switches and two WR nodes. Our current setup is shown in the attachment.
Switches:
HW version scb: v3.4 / backplane: v3.30
FPGA:LX240T; version: v4.1.2 (Benoit Rat); compiled at Dec 11 2014
16:27:35
Nodes:
wrc and wrpc-sw V2.1
Presumed that the grandmaster switch is powered-up well before the master switch, everything is nicely synchronized (PPS / TAI of grandmaster, master and both nodes).
For testing purposes we generate Ethernet traffic over the switches with two PCs running NetStress. Above ~500Mbps, master switch (switch 2) and the node connected to it (node 2) loose synchronisation with grandmaster: PPS outputs drift away with 4ns/s in one direction. Each node remains synchronized with its master, but apparently master-switch is no longer synchronized with grandmaster-switch. Once Ethernet traffic is stopped, synchronization can be achieved again by relaunching the PPSi of the master switch (web interface).