Hi,
We are using PTPd v1 and v2 and we see regular spikes in the offset. These occur every 9 hours 6 min and 30 seconds exactly. Normally the offset is within 1us or so but the spikes can take the offset to 100us to 2ms. This is observed in multiple locations that are not using the same grandmaster clocks so I don't believe the issue is there. Any ideas?
Regards,
Joel
Last edit: Joel Campbell 2013-01-14
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Hi Joel,
this period is near to 32767 seconds so maybe some integer arithmetic overflow. Can you tell more info about exact versions of PTPd and what is the grandmaster?
Regards,
Jan
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Can you identify if the spikes start with a one-way delay spike? Can you paste the timing log and syslog log from the moment of the spike?
32767 seconds is 9h 6m 7s, so close indeed. Older versions of ptpd had a sequence number rollover bug, but I'm not sure if it would cause an offset of the size you're seeing. All sequence numbers used by ptpd (at least v2) are 16-bit unsigned integers, so the range is from 0 to 65535. Signed 16-bit integers that would cause a rollover are not being incremented.
However this may be a sequence id issue - if you are sending 2 messages of any type per second, in that case the counter would reach the maximum value in 32767 sec.
Thanks
Woj
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Hi,
We are using PTPd v1 and v2 and we see regular spikes in the offset. These occur every 9 hours 6 min and 30 seconds exactly. Normally the offset is within 1us or so but the spikes can take the offset to 100us to 2ms. This is observed in multiple locations that are not using the same grandmaster clocks so I don't believe the issue is there. Any ideas?
Regards,
Joel
Last edit: Joel Campbell 2013-01-14
Hi Joel,
this period is near to 32767 seconds so maybe some integer arithmetic overflow. Can you tell more info about exact versions of PTPd and what is the grandmaster?
Regards,
Jan
I have both PTPd 1.0.1 and 2.2. Each site has a grandmaster that is the same version. The grandmasters sync to a NTP GPS device.
Joel,
Can you identify if the spikes start with a one-way delay spike? Can you paste the timing log and syslog log from the moment of the spike?
32767 seconds is 9h 6m 7s, so close indeed. Older versions of ptpd had a sequence number rollover bug, but I'm not sure if it would cause an offset of the size you're seeing. All sequence numbers used by ptpd (at least v2) are 16-bit unsigned integers, so the range is from 0 to 65535. Signed 16-bit integers that would cause a rollover are not being incremented.
However this may be a sequence id issue - if you are sending 2 messages of any type per second, in that case the counter would reach the maximum value in 32767 sec.
Thanks
Woj