[ntp:questions] Re: [Bug 177] Clock stepping messes up frequency. (fwd)

David L. Mills mills at udel.edu
Thu Oct 23 05:12:02 UTC 2003


Roy,

Thanks for that; it is exactly the kind of evidence I am looking for.
Read on.

Roy wrote:
...
> Hi Dave,
> 
> This may not be quite what you are looking for, but on my home NTP
> server I can frequently see times when the local clock drifts more
> than the step interval.  Here's a quick sample from an old loopstats
> file:
> 
> 52850 54400.148 -0.033933043 -119.678411 0.020886736 0.449823 14
> 52850 70785.151 -0.000764412 -119.426726 0.025568197 0.409380 14
> 52851 768.147 -0.054226675 -119.835474 0.034450925 0.409222 15
> 52851 17151.152 -0.027192308 -119.630382 0.034108466 0.368934 15
> 52851 49918.149 0.038115401 -119.381501 0.052961734 0.342885 15
> 52851 82685.150 -0.051950561 -119.724717 0.069032111 0.342968 16
> 52852 29055.149 0.009548464 -119.490629 0.067511066 0.319248 16

At this point things look fine. The frequency is wandering less than 1
PPM and the jitter, while significant, is not really large. Then the
nasty starts:

> 52853 70790.149 -0.303959474 -121.936574 0.161662943 1.253835 4

Where did that 300-ms spike come from? 1-PPM frequency error over about
40,000 seconds accounts for only 40 ms. I conclude most of that spike is
due to the network. Note that, at poll intervals above the stepout
interval (900 s) the clock filter and popcorn spike suppressor are less
effective. Even so, the frequency surge isn't very bad and things return
to nominal in a couple of polls.

> 52853 73715.141 -0.166563500 -178.880252 0.140650012 28.492537 4
> 52853 74620.152 -0.021400267 -129.140914 0.019528909 35.033826 4
> 52853 74684.148 -0.015529604 -129.200154 0.017238256 30.340198 6
> 52853 74748.147 -0.015106461 -129.257781 0.015280701 26.275398 6

Now, what I don't see here is a spike the other way, suggesting this is
an isolated instance, not a step that lasts awhile and that the
huff-n'-puff filter is designed for. Would you be so kind as to send me
both loopstats and peerstats for this or a similar interval?

Dave



More information about the questions mailing list