Re: about draft-arkko-multi6dt-failure-detection-00.txt

[Francis Dupont <Francis.Dupont@enst-bretagne.fr>]

 In your previous mail you wrote:

   >  - in 3.2 "Theoretically, it is also
   >    possible for hosts to learn about routing failures for a particular
   >    selected source prefix, even if no protocol exists today to
   >    distribute this information in a convenient manner."
   >    I disagree about the last part: there is a well known and simple
   >    protocol to announce routing failures for a source prefix: deprecate
   >    it, i.e., send RAs with zero preferred lifetime in the prefix info.
   
   In doing so you would be overloading two very different semantics on the 
   deprecated state:

=> I agree but this idea is very old (last meeting in Washington?) and
the wanted effect is the same.

     - this prefix will stop working in the future (sometime after the valid
       lifetime expires) so don't use addresses in this prefix for new
        communication; no effect on existing communication
     - this prefix is currently not working; immediately switch existing
       communication (as well as new communication) to use an address from a
       different prefix.
   
=> I disagree about the second semantics: the effect of deprecatin a
prefix does not include "immediately switch existing communication".

   Those two semantics are in conflict.
   
=> I agree if we want to extend the standard effect we have to add
at least an extra condition.

   So if we want the second semantics one would need to carry that 
   semantics. Of course, it wouldn't be hard to define some extension to 
   the router advertisements (and the router renumbering protocol if we 
   don't want to manually configure the routers) to say this; a single bit 
   in the prefix option might be sufficient.
   
=> IMHO we don't need a bit but simply to add a threshold condition, for
instance if the preferred time jumps from at least X hours to zero so
existing communications may be affected. And add an advice to use
smooth decreasing to zero in RAs for the standard semantics.

Thanks

Francis.Dupont@enst-bretagne.fr

PS: as far as I know this idea is not supported by IPv6 routers (nor Cisco
or Juniper at least). We developed locally a little tool (poll some
critical BGP routes by SNMP and reconfig routers) and it worked exactly
how we wanted: new connections used only source addresses in working
prefixes... I tried the extended semantics in my MIPv6 code many (6?)
years ago but without a good solution to the signaling protection issue.