[ntp:questions] Re: Improving ACTS performance
Richard B. Gilbert
rgilbert88 at comcast.net
Thu Apr 6 19:26:48 UTC 2006
Joshua Coombs wrote:
> I finally have my stratum 2 servers working as good as I think they can
> given their environments. Now I'm working on my stratum 1 clock which
> pulls it's time via ACTS.
> I'm currently dialing NIST so I can take advantage of the automatic
> latency compensation. Problem is, the time I get back seems to vary by
> up to 20ms. I may get 3 or 4 calls where the reported time stays within
> a couple ms of what I believe true time to be, and then a call will come
> back close to 20ms off. When ntpd has decided to stretch out to it's
> max call interval, that glitch induces drift in the clock, resulting in
> an eventual 40ms offset before it corrects.
> I've tried the following:
> - switched to manual latency adjustment using time1
> - only configured 1 number to avoid different latencies for different paths
> - dialing USNO instead of NIST
> I can't seem to shake the random wild offsets.
> I'm running the 4.2.1RC on FreeBSD 4.11, soekris 4501, and a Microcom
> 28.8 external modem.
> I'm open to any suggestions on how to improve things.
> Joshua Coombs
I've seen ACTS referred to as "Automated Containerized Trash Service".
What you see may be the best that ACTS can do.
Have you considered using another technology? A GPS receiver is not
terribly expensive (as low as $85 US) and such a receiver designed for
timing applications can deliver accuracy in the neighborhood of 1
microsecond. The PPS signal is accurate to better than 100 nanoseconds
but few computers are capable of that degree of accuracy. 1 microsecond
is easily obtainable. I use a Sun Ultra 10 (440MHz processor),
Solaris 8 and ntpd 4.0 and get that accuracy. That's eight year old
technology; more modern hardware and software might do better.
More information about the questions