[ntp:questions] Any chance of getting bugs 2164 and 1577 moving?
Richard B. Gilbert
rgilbert88 at comcast.net
Fri Mar 23 19:40:56 UTC 2012
On 3/22/2012 5:54 PM, David Lord wrote:
> unruh wrote:
>> On 2012-03-22, David J Taylor <david-taylor at blueyonder.co.uk.invalid>
>>> "unruh" <unruh at invalid.ca> wrote in message
>>> news:xLHar.38386$IQ1.34248 at newsfe18.iad...
>>>> Measure what? Why do you think that ntp reporting the offset with an
>>>> extra three decimal points would allow you to measure anything? What in
>>>> your mind would you expect to see in that output that would allow
>>>> you to
>>>> "measure" something that would tell you that the -19 was wrong?
>>>> ntpd DID measure something in order to determine that -19. What do you
>>>> think the extra decimal places would give you?
>>> Most likely I would be looking at a histogram of the reported
>>> offsets, and see whether it was gaussian, flat, or whatever, and how
>>> wide. I might learn something from that.
>> No. Not if it is just noise.
>>> Others have reported precisions better than -19, and also have a need
>>> for greater reporting precision.
>> That is a valid issue.
> I have servers currently with precision= of 18, 19, 20 but not
> scanned back in history more than today. The value varies, with
> temperature and system load which causes local temperature
> variations. The precision values vary and are just way points.
> With precision 20. I don't really need an extra decimal place
> but in a previous life was used to throwing away two results
> from five or more if from a greater number of samples.
> My standard pc hardware can't do any better.
>>> There seems to be an impression out there that I'm trying to show
>>> something is wrong - I'm not. I suggested an enhancement so that the
>>> precision of ntpq matched that of the loopstats. That's all.
>> precision is not accuracy.
>> In science we teach students not to report unwarranted precision-- the
>> precision should reflect the accuracy of the measurements. We keep
>> getting measurements to the mm and reported precision to angstoms
>> because that was what the calculator spit out. I am not averse to
>> reporting with a precion maybe up to a factor of 10
>> better than the accuracy, but any more is just silly and misleading (as
>> you are demonstrating in believing that a greater precision would convey
>> some extra information.
I remember this one from thirty or more years ago:
"Measure with micrometer, mark with chalk, cut with axe!"
More information about the questions