[ntp:questions] two machines syncing to the same master - hardware issues?
unruh at invalid.ca
Thu May 30 22:41:46 UTC 2013
On 2013-05-30, matthew.garman at gmail.com <matthew.garman at gmail.com> wrote:
> I have two servers in the same physical location that are syncing to the same two NTP servers. My understanding is that these NTP servers are in the same building as my servers, and we are connected via an in-building connection. In other words, network jitter should be reasonably low (at least compared to the Internet, for example).
> My two machines are somewhat different: "oldbox" is running NTP 4.2.2 as ships with RedHat 5.7; "newbox" is running NTP 4.2.4 as ships with RedHat 6.3.
> Oldbox has been used as our company's central NTP server for many years. But the machine is not dedicated to this purpose: it performs other business functions that are erratic in nature. So we procured newbox to act as a poor-man's dedicated NTP server. In other words, newbox does nothing but NTP.
> What we've found is that newbox generally has a much larger offset to its NTP peer, compared to oldbox. Likewise, newbox's offset also varies much more over time.
Plots would be good.
Plot the offset vs time from the peerstats file. You might discover that
your understanding is wrong.
> I'm measuring the offset via "ntpdate -q <peer ntp server>"
Very silly way of doing it, esp since your machine measures the offset
every few minutes already.
> I rebooted newbox yesterday, and it has taken about 20 hours for newbox's offset to be in line with oldbox's. And based on observation, I'm not so sure it will stay this way.
And? ntpd is slow. Always.
> Is it possible there is a hardware problem with newbox? I don't really understand NTP well enough to know where to start looking.
It is possible. What operating system is on newbox? And it could be that
something on the box is resetting the clock behind ntpd's back.
> I guess a related question is, what kind of offset can I realistically expect in such a situation? It seems like on oldbox the offset is consistently under 100 microseconds.
You can expect offsets of the order of 10-20microseconds IF the ntp
servers get their time from gps. If not, the servers themselves can be
woggling all over the place.
More information about the questions