[ntp:questions] Re: polling algorithm

David Schwartz davids at webmaster.com
Fri Sep 2 17:38:17 UTC 2005


"Brian Utterback" <brian.utterback at sun.removeme.com> wrote in message 
news:431861A4.8070108 at sun.removeme.com...

>>     Here's the simple proof. The time is 5:15. One server says:
>>
>> 5:14 plus/minus 2 minutes
>>
>>     The other says:
>>
>> 5:15 plus/minus 2 minutes
>>
>>     Now, what can the third insane clock say that will make you think 
>> it's any time other than 5:12-5:17?

> That's easy. How about 5:13 plus/minus 1 minute? Since you said the
> time is actually 5:15, this is by definition a falseticker, since
> the correct time is not within its interval. But the interval
> 5:13-5:14 overlaps all three servers, so this is where they
> all agree, so it must be where the correct time is, no?

    5:13 plus/minus 2 minutes would be correct. Only one clock claimed an 
accuracy window of one minutes. We can't base our conclusions on just one 
clock.

    It is possible that a naive algorithm could be mislead. However, it 
cannot be mislead by much. Remember, part of the definition of an "accurate" 
clock was that its confidence window was small. You've only thrown the clock 
estimate off by that much. So you've thrown it off by a small amount.

    Note that the correct time is 5:15 in the example. By definition, the 
two minutes the two clocks is off are a "small amount". So any time between 
5:13-5:17 would be acceptable, that is, off by only a small amount.

    DS





More information about the questions mailing list