[ntpwg] Issues with the NTP draft -06

[David L. Mills]

Brian,

The stratum is a red herring. It is the result of a comparison between 
synchronization distances, not the primal cause. As in all good Bellman 
Ford implementations, the distance is biased so as to minimize dither 
and resulting stratum flaps. While yours and other schemes might even be 
more reliable, I say for the umpteenth time the issue of backwards 
compatibility trumps all.

Dave

Brian Utterback wrote:

> One of the key pieces you need to remember is that the while the refid
> should be unique, unless the data you want to use for the server's
> refid is already in the data somewhere that arrives from the server,
> the refid must be derivable from the client.
>
> To answer Heiko's comment about the stratum level, the problem with
> stratum level is the propagation delay when the level changes. However,
> the reason having only two levels of loop detection works is that the
> stratum level generally works for larger loops.
>
> I have always wondered if there might not be a way to get better
> loop detection if you treated the refid as a sort of shift register.
> Each server would choose a random one byte number at shift that number
> on the the refid it passes to the client. Then the client would look
> at each byte of the refid from all of its servers. If the higher order
> byte matches the clients random byte, avoid it if you can. If the
> top two bytes match the lowest byte of the client's server's refid and
> the the client's own byte, reject it. Reject if more bytes match.