Re: ping-pong phenomenon with p2p links & /127 prefixes

[Ole Troan <ot@cisco.com>]

Jeroen,

>>> Unless you configure two /128's pointing to the remote side, which will
>>> then thus not be 'on-link for neighbor discovery, the little thing
>>> called the subnet anycast address will make sure that a /127 ptp simply
>>> does not work, unless you have a platform which disables the subnet
>>> anycast address of course.
>> 
>> It would seem disabling the subnet anycast is fairly widespread, then.
>> I have verified the use of /127 on several hardware forwarding platforms
>> from Cisco and Juniper. /127 works just fine, and prevents the ping-pong.
>> 
>> [One concrete example where /127 works: Juniper T1600 talking to Cisco
>> CRS-1 on an OC-768/STM-256 link.]
> 
> It is quite wide-spread indeed, and for instance Linux used to do it
> also until a kernel update in 2003 from 2.4.20 -> 2.4.21 and they
> finally implemented subnet anycast support(*) and suddenly it all
> started breaking as for IPng.nl at the time we used /127's and everybody
> with a Linux endpoint who did an upgrade of their kernels suddenly had a
> mysterious broken configuration.
> 
> Thus, do ask Cisco and Juniper and other vendors where this now 'works'
> if this intentional, or if they might finally comply to the IPv6
> specifications one day, as then you might better watch out for this as
> it will break your network. For the vendors that have it, it might maybe
> be an idea to have a 'disable subnetanycast' command or similar so that
> one can explicitly mark a prefix that way.

it is intentional.
there is a command to enable support for subnet-router anycast if use of that is desired.

is there _any_ operational experience with the use of the subnet router anycast address?
asking the question another way. is it still a good idea, or was it ever?

(for 6rd we have explicitly required support of it, for discovery of the real BR unicast address, since we're using IPv4 anycast).

cheers,
Ole