> The _sore_ thing is that with 2.4.x-nano kernels offsets/jitter got > that good within minutes. Now many years later it takes days... :-(
I think that's a bug introduced by the tickless scheduler work. > For my daytime use of ntp, convergence in less than 5 minutes is > essential. Due to some unusual peripherials, that often have drivers > dictating a certain kernel version its is often impossible to find a > linux kernel version that support both the proprietary drivers and > decent ntp-server performance. Do you build your own kernels? It looks like there is a simple change that fixes the convergence. http://lkml.org/lkml/2009/5/4/373 There is also the TSC calibration mess. If you are only running on one machine, or one type of system, you can patch the kernel to use a hard-coded answer. Or you can try one of the non-TSC clock sources. -- These are my opinions, not necessarily my employer's. I hate spam. _______________________________________________ time-nuts mailing list -- time-nuts@febo.com To unsubscribe, go to https://www.febo.com/cgi-bin/mailman/listinfo/time-nuts and follow the instructions there.