[ntp:questions] Bug 2341 - ntpd fails to keep up with clock drift at poll>7
snews at lordynet.org
Tue Dec 3 13:39:14 UTC 2013
> On 2013-12-03, David Lord <snews at lordynet.org> wrote:
>> unruh wrote:
>>> On 2013-12-02, David Lord <snews at lordynet.org> wrote:
>>>> Martin Burnicki wrote:
>>>> My server with Sure gps/pps has offset below 3 us other than when
>>>> nightly cron jobs give a couple of 35 us spikes. From loop_summary
>>>> over 7 days, typical range for rms offset is 3.9-6.1.
>>>> Over 7 days my four pool servers have following rms offset ranges
>>>> ntp0=784-1646, ntp1=405-837, ntp2=310-434, ntp3=586-1270 but there
>>>> were numerous reboots and ntpd restarts over that period.
>>>> I want to try using a stable external system clock source, TCXO,
>>>> OCXO or rubidium laser.
>>> A sure gps IS a stable external system clock source. What more do you
>> ? it is a PPS source not a system clock source.
>> I'd prefer the 35 us blips to go away which would require the PC
>> to be idle 24/7 or a use of a stable external system clock.
> ??? That 35us blip is quite possibly there either because of delays in
> reading the clock, which would also be delayed with an external clock.
> Ie, I would need to see some evidence that an external clock would be
> any more accurate in getting the time into the system, than would the
>>> And those offsets for your pool servers really seem very high.
>> did you miss that:
>>>>> were numerous reboots and ntpd restarts over that period.
> No I meant that delays I have seen getting time from stratum 1 servers
> were in the tens of microsecond not hundreds or thousands of microsecond regime. Now it may be that the
> pool servers you are using are really just horrible as clocks
> (milliseconds I call just horrible). But I certainly would not expect
> network delays to give that kind of errors.
> Also, I would assume that you are not taking the behaviour during
> startup transients into account.
>> but peak offsets are still 300-500 us and I'd prefer if they
>> were even lower, again requiring a more stable system clock.
> Why would they be lower. Those are probably the fluctuations in teh
> remote clocks, and have nothing to do with the fluctuations in your
> local clock.
>> What do you think the local offsets should be for a pool server?
> It depends on the pool server. Some simply use nmea and claim to be
> stratum 1. As I say, the offsets from a reliable pps driven stratum 1
> from across the country are in the 10s of microsecond range for me.
> (even though delays were in the 10s of ms range)
Whilst on adsl delays for me have increased from about minimum
of 5ms to current minimum of 22ms. Distance from telephone
exchange has increased from about 400m to about 1400m
Ntp2.lordynet.org.uk has a local stratum 1 source almost 1ms
distant and 5 x stratum 2 sources with delays from 22ms to 45ms.
Other three servers have internet sources mostly at stratum 2
but some at stratum 3. That means my four pool servers are at
stratum 2, 3 or 4. Local offsets of the servers are mostly
< 500us. I can't imagine ntpd or even chrony achieving 10s of
microseconds offset from a remote stratum 1.
I might just try chrony on a spare pc, I used chrony when I had
a dialup connection to Demon along with ntpd for the lan.
More information about the questions