[Fwd: Re: [ntp:questions] ntpd, boot time, and hot plugging]

Richard B. Gilbert rgilbert88 at comcast.net
Thu Feb 3 01:59:55 UTC 2005

Kenneth Porter wrote:

>Brad Knowles <brad at stop.mail-abuse.org> wrote in 
>news:mailman.8.1107364517.583.questions at lists.ntp.isc.org:
>>     Please note that ntpdate is being deprecated for a reason.  It 
>>does *not* give you good time sync.  Moreover, it does not *keep* the 
>>server in good time sync.
>Can you point me to anything that talks about what "good" means here, and 
>how good is good enough? Does the average web/mail server need sub-
>millisecond accuracy? It can still benefit from regular corrections. What 
>are typical application requirements for time quality?
ntpdate sets the clock on a one time basis.  It does not correct your 
clock frequency so whatever error there is will keep right on causing 
your clock to gain or lose time.   It unconditionally sets the clock to 
the time supplied by the server which may or may not be correct!

A better way is to use ntpd and start it with the -g option which will 
cause ntpd to set the clock on a one time basis but then continue to run 
and discipline your clock to bring it into synchronization with the 

Do you need sub-millisecond accuracy?  Only you can say.   Using only 
publicly available network servers, you probably will not get 
sub-millisecond accuracy!  If you have good servers and good network 
connections to them you might reasonably hope to be within ten 
milliseconds or even five milliseconds of the correct time.  If you use 
a hardware reference clock, such as a GPS receiver you can get 
sub-milliecond accuracy; even sub-microsecond accuracy. 

Having correct timestamps on mail headers can be most helpful when it 
becomes necessary to trace mail back to its origin.  It's also helpful 
if you are trying to debug something and all the systems involved have 
the same time; when a log file says that something happened at 
13:07:33.48 you can believe it and you can make meaningful comparisons 
with other systems log files.

More information about the questions mailing list