[ntp:questions] Re: Testing SNTP devices

Terje Mathisen terje.mathisen at hda.hydro.com
Thu Aug 28 08:55:03 UTC 2003

dave wrote:

> I have the job of testing some remote utility meters that use SNTP to
> sync their clocks.  They can't do NTP and the customer is all up in
> arms about it.
> Customer is overly concerned about accuracy, even though 1 second is
> more than good enough for this application.
> The meters will be fed from a known good NTP server, on our side of
> the firewall.
> The customer want us to tell him the clocks are within xxx msec.  What
> is the best way?
> Packet sniff on port 123?  
> I am in a windows environment.

The sniffer idea is good:

If the rount-trip delay, as measured at the SNTP device network 
interface, is N ms, then N should be a reasonable estimate of the 
maximum error of said device.

However, if the internal clock of this device is badly wrong, and the 
SNTP implementation doesn't try to correct for first-order drift errors, 
then you have basically no way to estimate how large the offset can get 
between polls of the upstream NTP server.

I.e. the clock error will follow a sawtooth pattern, with gradually 
increasing offset between each SNTP poll cycle that yanks it back.


- <Terje.Mathisen at hda.hydro.com>
"almost all programming can be viewed as an exercise in caching"

More information about the questions mailing list