[ntp:hackers] NTPv4 Brian Version

David L. Mills mills at udel.edu
Thu Aug 11 19:59:34 UTC 2005


H,

I don't seem to be getting through. It once worked; subsequent requests 
did return different first entries. At least now, the same first entry 
is returned and it doesn't rotate. However, with dig and nslookup 
subsequent requests mostly do return different first entries. Sometimes 
two queries return the same first entry, but that is expected with 
statistal ordering.

The code really should use all entries in the reply subject to the 
configured sys_maxclock value, but I have no idea how to do that and 
preserve compatibility.

Right now, subsequent duplicate first entries are treated like remote 
configuration message and are tossed. With asynchronous resolver, 
duplicates would have to be filtered out.

I have updated the manycast algorithms to conform with how I expect the 
poll to operate with appropriate resolver. In the transmit() routine in 
ntp_proto.c the resolver would be called in the same way the manycast 
client message is sent; that is, the resolver would be called every 64 s 
as long as the preemptable associations were less than sys_maxclock. The 
IP addresses found would call newpeer() as in the manycast server stub. 
Th preemptable associatinos would be purged at the same time the Autokey 
server seed is refreshed, about once per day. The associations have to 
be purged because the cookies have to be refreshed.

Dave

Harlan Stenn wrote:

> Dave,
>
> This is because ntp only pays attention to the first IP returned in a
> DNS resoltion. If the resolver code does not "rotate" answers, multiple
> queries will return the same IP.
>
> The feature we need is a way to ask for up to N IPs from a single name,
> and we need to DTRT WRT the 'restrict' lines.
>
> This is what I wrote in the URL I just posted to this thread.
>
> H




More information about the hackers mailing list