[ntp:questions] Question on "calldelay"

Uwe Klein uwe_klein_habertwedt at t-online.de
Mon May 7 07:37:11 UTC 2007


Danny Mayer wrote:

> It's probably not wise since if the connection is not up and ready, then
> you are just wasting the packets you send.

ALL IPs ARE STATIC!
the setup is:
<ntp server> ..<some inhouse routers/switches>..<ciscoA> ....<ISDN leased line> ...<ciscoB> ... <my 
ntp client>

ciscoB log messages go to the ntp client machines syslog.

In a succesfull exchange I see 8 udp packets (calldelay15 + burst):

ReqC --> AnsS
..<15seconds delay>..
ReqC --> AnsS
..<~2sec>..
ReqC --> AnsS
..<~2sec>..
ReqC --> AnsS

In an UNsuccesfull exchange I see _1_ packet (calldelay15 + burst):

ReqC -->
<..76 seconds timeout, dialup taken down. >

> I suspect that the real problem is that the box is getting an IP address
> assigned by DHCP and you need to rebind the local address first so that
> the packet can go out from a valid address.
there is no  IP change via DHCP or otherwise involved.
> 
> Did NTP rebind the local address first?
NO address change involved.
> 

> The protocol doesn't care. This is UDP. If it doesn't get a response it
> will try again later. When later is depends on the stage that it's at.
Right udp per se does not care.

what about ntpd:
at what moment in the exchange is the calldelay rundown started ?

imho it should be:
	ReqC0----------------15sec>ReqC1
		..AnsS0			..AnsS1

my current guess is :
	ReqC0->AnsS0----------------15sec>ReqC1
						..AnsS1

I have looked into ntp_proto.c where this is handled
but am currently not through understanding the state stepping.

uwe




More information about the questions mailing list