[ntp:questions] NTP + PPS on Atom motherboards
david-taylor at blueyonder.co.uk.invalid
Wed Aug 1 10:17:45 UTC 2012
On 01/08/2012 09:46, Hahn, Ron wrote:
> I have been working for some time now trying to get four different Asus Atom motherboards to successfully work as a NTP stratum 1 server. I am using the FreeBSD 8.3 OS with PPS compilation in the kernel, and the ntp-devel port (4.2.7p2?? version ntp). Thanks to the David Taylor web site, I am using the Sure GPS boards to drive these motherboards and also his MRTG script to monitor these things too.
> What I am seeing is that the time is off by almost 53,000uS on the MRTG graph and almost 500mS sometimes on the outputs of ntpq -c pe at times. Other times it is a few uS from PPS. I have been seeing this on three different boards (D525, D510, and 330 CPU) so this is repeating. I was thinking this might be the tty port so I have tried both the tty ports and still the same.
> I have used exactly the same recipe on a Core 2 Duo Tyan server and the times are maximally off by only +4uS/-2uS from PPS. So I am thinking there is something fundamental wrong with the Atom boards. I have repeated the experiments with a Pentium 4 server in another location and I am also seeing excellent timekeeping too. I am thinking that maybe the Atom clock on the board is too consumer for these uses??
> Has others experienced these difficulties with Atom motherboards as Stratum 1 servers?
This is the performance I see from an Intel Atom system:
about as good as I would expect. The motherboard in that PC is an Intel
D410PT. So I think that the problem is not from using an Intel Atom per
se. I also have an ASrock Intel system which is LAN-synced, and the
offsets there are a few milliseconds, but not 53 ms.
More information about the questions