Has ptpd been tested with the 2.6 kernel? I've been running some tests on multiple systems using the 2.4 kernel (RedHat 9 out of the box) and it works pretty well.
However, running on systems using the 2.6 kernel, the master to slave offset (as well as the master to slave delay) swings violently back and forth between a very large positive and negative value every few minutes (the swing is on the order of minutes). Looks like some sort of limit-cycle behavior.
What is interesting is if I run the master on a 2.4 system and the slaves on 2.6 systems, everything is fine. The master doesn't like to run on a 2.6 system. Any ideas?
Thanks,
Gerry
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
I run PTPd on my Linux 2.6 machines. The 2.6 kernel is less responsive than the 2.4 kernel, but latencies are usually on the order of microseconds to milliseconds.
Offsets larger than one second are not run through the servo, so I don't think there's anything in there that could cause this. To slew the clock by a second would take a *very* long time even at the maximum frequecy adjustment. When off by more than a second, PTPd will try to reset the clock. If PTPd cannot reset the clock well within a second, it's pretty much SOL.
It might be worth looking into some of the tools that the real-time linux projects are using to porfile kernel latencies.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Has ptpd been tested with the 2.6 kernel? I've been running some tests on multiple systems using the 2.4 kernel (RedHat 9 out of the box) and it works pretty well.
However, running on systems using the 2.6 kernel, the master to slave offset (as well as the master to slave delay) swings violently back and forth between a very large positive and negative value every few minutes (the swing is on the order of minutes). Looks like some sort of limit-cycle behavior.
What is interesting is if I run the master on a 2.4 system and the slaves on 2.6 systems, everything is fine. The master doesn't like to run on a 2.6 system. Any ideas?
Thanks,
Gerry
I run PTPd on my Linux 2.6 machines. The 2.6 kernel is less responsive than the 2.4 kernel, but latencies are usually on the order of microseconds to milliseconds.
Offsets larger than one second are not run through the servo, so I don't think there's anything in there that could cause this. To slew the clock by a second would take a *very* long time even at the maximum frequecy adjustment. When off by more than a second, PTPd will try to reset the clock. If PTPd cannot reset the clock well within a second, it's pretty much SOL.
It might be worth looking into some of the tools that the real-time linux projects are using to porfile kernel latencies.