[ntp:hackers] Cool new stuff

David L. Mills mills at udel.edu
Fri Jul 21 14:21:40 UTC 2006


The interface code is cool, but does it have to scan for changing 
loopback and refclock interfaces?


Frank Kardel wrote:

> David L. Mills wrote:
>> Guys,
>> We are in the final stages of testing radically new stuff:
>> 1. Remote configuration/mobilization/demobilization at any time using 
>> ntpq. The commands are identical to those in the configuration file 
>> now. In principle, it is possible to bring up a raw daemon and 
>> configure it entirely from ntpq. 
> a very welcome feature!
>> As you might expect, the configuration code is almost completely 
>> gutted and the ntpdc remote configuration no longer works. On the 
>> other hand, al lot of things that were once applicable to all 
>> associations can now be applied on a per-association basis. At this 
>> time the command syntax and semantics support the current 
>> interpretation. The plan is to provide a command line switch to 
>> enable the legacy interpretation and whatever new interpretation 
>> seems advised. This is expected to evolve using C-type syntax with 
>> curly brackets and semicolons. 
>> The syntax change will be considered radical by many and the end of 
>> ntpdc remote configuration will be considered evil by some. The ntpdc 
>> program continues to be supported but without the capability for 
>> remote configuration.
> This this is finally the end that we need to tunnel parameters through 
> time1/2  and the flags. Finally we can have proper syntax for the
> implemented semantics. Good!
>> Lots of things haven't been thorougly tested and several 
>> improvements, like the ability to read configuration files by ntpq, 
>> need to be on the todo list. However, Sachin Kamboj, the ace 
>> implementer of this stuff, is no longer supported by contract and is 
>> volunteering his own time. Probably the best way to proceed is to 
>> bring up a parallel version for a very short breaking-in test and 
>> then switch it to ntp-dev. Sachin is now doing this with diffs, which 
>> simplifies the switchover. I'd like the autoconfigurus to look at it 
>> and see if the syntax tree construction and merge can be automated.
>> If you have thoughts about this, please honk. When ready for test I 
>> will aadvise. Probably a ntp-new directory on deacon:/backroom.
>> Dave
>> P.S. Testing has been complicated by intermittent huge spasms of 
>> debug messages apparently emitted by the interface code. Can this be 
>> attentuated or moved to a higher -d level? DLM
> will move to a higher debug level (like >= 3) and remove some output. 
> These spams are generated by the other
> cool new stuff that makes ntpd cope with changing interface 
> configurations and thus enables it the be a well
> behaved citizen in the mobile world or cope with "Zwangsgtrennung" 
> (sick concept by ISPs for you to get forcibly a new
> IP address).
> While talking about debug levels shouldn't we  consider  classifying 
> debug  levels
> (maybe like the logconfig command)  so we can selectively crank up  
> debug for io/loopfilter/clockselection/crypto/peers ?
> I have basically the same problem when debugging the i/o engine - I 
> get swamped by all the other areas of the code.
> Frank 

More information about the hackers mailing list