Re: drvTS/NTP issues?

[Noboru Yamamoto <noboru.yamamoto@kek.jp>]

Hi,

Peregrine McGehee wrote:
> 
> We have an IOC configured as a soft timing slave with no
> master IOC on the subnet so it is syncing with our NTP
> server once every 10 seconds.
> 
> While the reported correction factors are always reasonable
> - of order 0.0003 - and we see the difference between the
> IOC and NTP time recover the 0.03 seconds lost every 5 minutes
> when looks again for a master timer, we find that the IOC
> will sometimes and suddenly fall behind in chunks of about
> 10 seconds - and never recover. The IOC has been as much
> as 40 seconds behind the NTP clock.
> 
> Has anyone ever seen something like this before? Could it
> be significant that the quantum of lost time is about the
> same as the NTP sync rate?

1) As far as I know, standard TimeStamp subroutines give up to
synch clock to the NTP master if the time difference is too large.

2) Standared algorithm can correct offset between IOC and NTP master.
However, if IOC clock rate is slightly different there will remain
constant 
offset between the IOC clock and NTP clock. We use the modified version
of Timestamp subroutine for this( written by myself).  
It works on EPICS3.13.2. But I didn't test it with latter version of
EPICS, yet.

3) On a few of our IOC, we still encounter the large clock offset sometime.
I'm still looking at this problem. 

> 
> We are running EPICS R3.13.1 on MVME-162s (old style and
> Petra) and vxWorks 5.1.1.
> 
> Aloha,
>         Peregrine
> --
> Peregrine M. McGehee            (505) 437-6822
> Sloan Digital Sky Survey        Apache Point Observatory

Noboru Yamamoto
KEKB control group
KEK, JAPAN