[ntp:hackers] Mitigation rules

David Mills mills at udel.edu
Mon Apr 6 18:27:40 UTC 2009


Warner,

I don't know of any configurations that are backwards incompatible, 
although some folks have managed to configure a network in some bizarre 
ways, especially using multiple local clock drivers. Most of the changes 
had to do with strict precedence considerations under various fallback 
conditions and the behavior when no sources are available. Some were 
rather obscure bugs, like using the combining algorithm when multiple 
orphan servers are available and allowing the prefer peer to override 
orphan mode.

There remains a bug seen only with Solaris. The distance metric for 
orphan mode is the IPv4 orphan server address or the hash of the IPv6 
address, which works find for FreeBSD. For some reason the value 
returned for any address in Solaris is the same. This should cause 
trouble when looking for loopback errors, not just orphan servers. So 
far not resolved.

Dave

M. Warner Losh wrote:

>In message: <49DA35A7.7030209 at udel.edu>
>            David Mills <mills at udel.edu> writes:
>: Guys,
>: 
>: NASA/JPL has decided to implement the NTP interleaved modes for Mars 
>: orbiters and deep space missions. These modes are implemented in the 
>: current development verion. This requires some rethink on the mitigation 
>: rules and prefer peer concepts as the intervals between updates can be 
>: significantly longer than the current engineering model. After some 
>: thought I have revised the mitigation rules to support such things as 
>: multiple prefer peers, multiple PPS sources and engineered fallback 
>: should all sources fail. The revised rules are intended to cover the 
>: existing GPS/PPS paradigm, sole PPS source and other more or less 
>: bizarre hookups. The revised rules are at 
>: www.eecis.udel.edu/~mills/ntp/http/prefer.html. They are in fact 
>: implemented in the current development version, which should in almost 
>: all cases be backwards compatible with previous versions..
>
>Cool Stuff David.
>
>"Almost all" cases.  Which ones aren't backwards compatible?
>
>Warner
>  
>



More information about the hackers mailing list