[ntp:questions] Re: System clock

SivaKumar Subramani ssubrama at ssd.usa.alcatel.com
Sat Oct 8 07:10:28 UTC 2005


David,

The requirement is to run a separate clock by the NTP process, which has all the
capability of system clock processing ( Simulator of a system clock all together).
As you very clearly said, we need to modify the kernel, by writing wrapper to
system call functions or replace entire kernel calls w.r.to the system time
management calls to implement the functionality of the system clock. This shall be
in the user space program.

Our architecture running in a heterogeneous OS network, like our processors is the
ONLY one running solaris OS and the other procs are running on different OS like
cekos,lynxOS...etc.,. The TOD distribution program running in LynxOS...this has to
be retained as it is without disturbing the existing architecture. As you said the
requirement specification itself looks like complicated, but we have to go with
it.

Any comments on this requirement further. I hope the requirement is very clear
now.I'm looking for suggestion and design input to implement system clock
processing in the user space program.

Thanks in advance for your  design inputs.

Thanks
Sivakumar

David Woolley wrote:

> In article <43467245.644D67E3 at ssd.usa.alcatel.com>,
> ssubrama at ssd.usa.alcatel.com (SivaKumar Subramani) wrote:
>
> > Actually my requirement is to run a clock  which can be controlled by NTP
> > daemon process. I  want NTP to manage the clock functionality to do on my
> > clock not on the system clock.
>
> In theory this could be done by accumulating the phase correction that
> would have been accumulated on a properly controlled clock (including
> interpolating the effects of kernel PLLs, etc).  But see below.
>
> >                               Actually a virtual clock shall be maintained by
>                                                          ^^^^^
> > the NTP daemon process with system clock functionality. Because the system
>
> "the NTP daemon" obvious has a special meaning in this specification,
> we need to know what that is as it doesn't have the meaning that would
> normally be understood; whoever wrote this, rather legalistic,
> specification has no authority to specify that's behaviour.
>
> The general style of this specification looks like it was written by
> someone who felt they had to justify themselves as a specification writer,
> or was written by a competing bidder in order to load the requirement in
> favour of their solution.  (I've seen both over specification through
> inexperience and overspecification by a consultant whose bidding.)
>
> > clock shall be controlled by some other program in the network, this shall
> > distribute the TOD to the entire network including my processor. This TOD
> > clock sometime may be set by the user manually.This TOD process shall get the
>
> This is a killer.  It means the only way that you can meet the requirement
> is to completely duplicate the software clock processing; that can't be
> done within a user space program like ntpd, it can normally only be done
> by modifying the kernel source code.  To track the accumulated correction,
> you have to assume that no corrections of any kind are being applied.

Yes, you are correct. By modifying the kernel source only this can be satisfied.
How to implement the system clock capability in the user space program like
ntpd.Any device driver required to implement this functionality of the system
clock.Just started preparing on this line. Any inputs or alternate design is
solicited.
Why don't we think in NTP protocol itself to implement this. As per user request,
there shall be some option in the config file for NTP to behave like setting
system clock or daemon clock.

I've one more doubt on the nanokernel source, what for this source is used?

>
>
> Again, we have to ask:  "what is the real requirement?", and by that we
> don't mean what the specification document says but the reasons that
> forced the specification document to say them.
>
> > I've attached the architecture diagram to make it further clear on the
>
> EIther you didn't attach it or it got stripped by the gateway.  This
> is primarily a newsgroup and it is a discussion newsgroup, so binaries
> would not have been acceptable.
>
> _______________________________________________
> questions mailing list
> questions at lists.ntp.isc.org
> https://lists.ntp.isc.org/mailman/listinfo/questions




More information about the questions mailing list