[ntp:questions] Query about NTP accuracy

Richard B. Gilbert rgilbert88 at comcast.net
Fri May 22 12:47:52 UTC 2009

Andy Yates wrote:
> Richard B. Gilbert wrote:
>> Andy Yates wrote:
>>> Does anybody have any figures that shows the effect on accuracy of an
>>> NTP v3 client using a stratum 1 server rather than a stratum 2 or 3
>>> server? It's all in a GE LAN based scenario, commercial stratum 1
>>> servers connected to GPS and stratum 2 and 3 servers are typically
>>> dedicated Linux boxes.
>> Stratum has very little to do with accuracy!  It's simply a count of the
>> number of servers between you and the root of the tree; e.g. the atomic
>> clock.
> Many thanks Richard - exactly my argument too.
>> If you have low latency connections a stratum three server can give you
>> millisecond or maybe microsecond accuracy.  Let those connections get
>> really busy and your accuracy goes down the toilet even at stratum two!
>>> The reasons is that I would rather scale by adding strata - its a very
>>> big data center with thousands of clients and has several "zones" that
>>> are isolated. However some opinion is suggesting we run IRIG-B between
>>> the GPS receiver and a bunch of stratum 1 servers and clients access
>>> these directly. Much more expensive and any increase in accuracy from a
>>> client experience may be negligible.
>>> However I'm been pressed to supply an SLA for accuracy. My argument is
>>> that although you can get your stratum one server to synchronize to
>>> microseconds of UTP, as soon as the client uses NTP v3 over the LAN,
>>> even a GE LAN, then the accuracy degrades and putting well designed well
>>> specified stratum between the boxes is not going to decrease accuracy
>>> sufficiently to warrant purchasing many stratum one appliances.
>>> Thoughts?
>> NTP v4 over the LAN would be just as bad as V3. Think about broadcast or
>> multicast!  That way you don't clutter up the net with queries and
>> response.  Instead, every N seconds your server announces: "at the chime
>> the correct time. . . ." and your clients can adjust their clocks as
>> necessary.  Your broadcast and/or multicast clients will exchange
>> packets initially in order to establish the round trip delay.  After
>> that initial volley, the clients just sit and listen.
> Agreed - we already use multicast clients by default however my point is
> that having your stratum 1 server synchronised to microseconds does not
> guarantee a microsecond NTP client experience and adding additional
> strata does not materially effect this. However I need evidence to back
> this up to support the SLA - I don't want to get into having to get
> testers in to prove the obvious however if anybody has already looked
> into this and has evidence then this would be useful.

I'm not sure that ANYTHING, other than a local atomic clock, is going to 
  get you microsecond accuracy.  10 milliseconds is probably attainable 
for other than Windows clients.  The Windows clock ticks every 17 
milliseconds and may be accurate to within a few hundred microseconds 
but, given the available resolution, you have no way to determine if it 
is that accurate.

A GPS timing receiver may provide a Pulse per Second output and 
guarantee that one edge of that pulse is within 50 nanoseconds of the 
"top of the second" but getting that signal into your computer with that 
accuracy is not easily done and may not be possible at all!

Computers keep time as an "afterthought"!

More information about the questions mailing list