[ntpwg] Network Time Protocol (NTP) Options for DHCPv6

[Danny Mayer]

Benoit Lourdelet (blourdel) wrote:
> Hello,
> 
> A possible usage of all the NTP parameters specified in the new options
> is to configure router gateway . In that scenario, the goal of a Service
> Provider or an Enterprise IT  is to autoconfigure the gateway as much as
> possible, including some NTP client behavior, in order to keep the
> control. An additional benefit is that the gateway can be shipped to
> remote sites virtually without any configuration while keeping full
> control of the NTP behavior.
> 
> Following that line of thought, I may still see a need for  "minpoll,
> maxpoll, I and B fields".
> 

No you don't, as I stated elsewhere, iburst should always be used by
default but neither burst nor adjustments to the minpoll and maxpoll
values should be made except under very special circumstances. burst is
extremely unfriendly to the server and minpoll and maxpoll can make the
algorithms used in NTP less stable. You really have to be an expert in
NTP and its algorithms to change these from their default value.

> Considering a central DHCP server, the use of remote-id [RFC4649] or
> just link-id may tell the server from where 
> the request is coming then giving it the knowledge of the delay to
> apply.
> In the case of a DHCP server hosted by the first hop, the DHCP server is
> well positioned to know the delay.
> 

That's not true. The delay is between the NTP client and the NTP server
and includes delays of any intermediary points like a router.

> Considering the current option format which includes more than IP
> addresses, the choice of multiple occurrences of the 
>  same option looks sensible. Should we reduce the new options to IPv6
> addresses only, a list makes more sense.
> 

No you should be using a DNS name and not IP addresses. The client then
can use the DNS to lookup the name and get a bunch of IP addresses. The
new pool option will then configure all of them (up to 10) for NTP servers.

> 
> Benoit