[evla-sw-discuss] NTP, EVLA MIB's and phase jumps

[James Robnett]

    I've now returned NTP to the state it was this morning

Amos and Andy,  our two NTP servers both point to our two
stratum-1 hardware clocks as servers and to each other as
peers.

They also point to our third NTP server (which does not
multicast) as another peer.   This is to help prevent issues
on reboots.

The only *net* change I've made is to set all 3 to use acpi_pm
rather than TSC for calculating their internal clock frequency.

All 3 servers are reporting times that I would take to mean
we shouldn't see phase jumps.

Sadly it takes time to 'fix' the frequency and all 3 are
still slowly slewing to a wrong time.  I'm hoping the
correction is faster than the error and they'll slowly
correct to +/- 5 to 10ms in the next few hours.

Amos, Andy and the 3rd server are at +/- 28ms, 15ms and
24 respectively.  Earlier they were anywhere from 40 to 120ms.

I'll continue to keep an eye on it.

James
James Robnett wrote:

> 
>     I'm going to watch it for a bit longer and then restart
> ntp and let it return to multicasting./
> 
>     I'm also going to apply the fix to amos right now.
> 
>     I don't *know* that this was the problem or that it fixes
> it but early indications are that it does.
> 
> James