Brian Utterback <brian.utterb...@sun.com> writes: >Unruh wrote:
>> You misunderstand. I was not (necessarily) asing for increased limits on >> the slew. Your proposal requires a rewrite of ntpd which allows that >> extra knob to be inserted into the ntp.conf file, to be read by the >> program, and then implimented by the program. My suggestion is that that >> knob be automated by ntp itself. After all your knowledge of how much of >> a correction to the rate is needed in adjusting that knob comes from the >> evidence supplied by ntp itself. What I am saying is that that evidence >> should be collected by ntp itself and ntp itself should twiddle that >> knob, not stick in the extra step of having a human try to evaluate the >> output of ntp, determine how much that knob should be set, and then >> setting it. Remember the tickadjust is very crude. It jumps in amounts >> of 100PPM ntp should be able to, extremely rapidly, determine which of >> the 100PPM bins to set the tickadjust to and then leave it. It should >> not require human intervention. This would then leave the full 500PPM >> for adjusting offsets and responding to rate changes caused by thermal >> effects for example. As it is now, that 500PPM can be largely eaten up >> by compensating for the constant large drift rate of the clock. >Absolutely, although at some point even the coarse knob ought to have >limit beyond which it is time to call in the marines. But NTP can take >quite a long time to settle down when the drift is very high, so >getting it down to under 100ppm would help a lot. As odd as it sounds, >I might be inclined to go back to having a separate start up program >(like ntpdate was) to quickly set the date and drift. Having to get >the date to within 64 years is another one of those odd, must be done >outside the protocol, quirks. Ah yes, ntp's long convergence time. Within one second you can determine the drift rate to better than 100PPM ( that is 100ms/s) Ie, you can go out and query your ntp source and 10 seconds later query it again and set the coarse drift rate. There is nothing rocket science in that. Then let ntp take its 10 hours to correct the remaining 100PPM. Since the whole point is that this is a intitiation effect only, you do not have to worry about the effect this will have on the network. I agree that if the clock is out by 100,000PPM or more you have other worries. >Brian Utterback _______________________________________________ questions mailing list questions@lists.ntp.org https://lists.ntp.org/mailman/listinfo/questions
