On Dec 13, 2012, at 4:13 PM, Peter Psenak wrote:

> Hannes,
> 
> On 13.12.2012 15:57, Hannes Gredler wrote:
>> 
>> On Dec 13, 2012, at 3:31 PM, Peter Psenak wrote:
>> 
>>> Hannes,
>>> 
>>> On 13.12.2012 15:21, Hannes Gredler wrote:
>>>> 
>>>> On Dec 13, 2012, at 12:43 PM, Stewart Bryant wrote:
>>>> 
>>>> [ … ]
>>>> 
>>>>>> can we say that the PQ node address used for targeted LDP session is 
>>>>>> selected in following order of preference:
>>>>>> 
>>>>>> 1. PQ node OSPF router-id, if it is advertised as /32 prefix by the PQ 
>>>>>> node itself
>>>>>> 
>>>>>> 2. Highest /32 address advertised by PQ node in it's Router LSA
>>>>> Why do we need (2). What do we do if an interface cycles?
>>>>> 
>>>>> Surely the most we should say is that we SHOULD establish a TLDP session 
>>>>> with the
>>>>> OSPF router-id, if it is advertised as /32 prefix by the PQ node itself, 
>>>>> else any
>>>>> other IP address for the node may be used.
>>>> 
>>>> other implementations may not be willing to accept T-LDP sessions on 
>>>> non-loopback
>>>> adresses.
>>> 
>>> how do you suggest to distinguish between /32 belonging to the loopback 
>>> from /32 belonging to other interface types?
>> 
>> for OSPF traffic engineering deployments:
>>  - if a /32 stub route advertisement of a type-1 LSA matches the TE router 
>> address TLV in the type-10 LSA
> 
> are we only allowed to advertise loopbacks in the TE TLV?

remember, in lieu of explicit signaling the purpose of this is finding the 
"most likely" transport IP for the T-LDP session.
Those are heuristics and as such imperfect. If i can cover a high percentile of 
the deployed base, then 
i usually can convince operators to fix their broken configs where the 
router-id/TE-router-ID does not match
the routable transport loopback IP.

>> 
>> for OSPF non-traffic engineering deployments:
>>  - if a /32 stub route advertisement of a type-1 LSA matches the router-id 
>> in the LS header
> 
> does not work if the router-id is not advertised by the node as a stub-link, 
> which is a valid case.

perfect, so lets declare this a "broken" configuration for automated T-LDP 
bringup w/o explicit signaling.

i do sense some discomfort with the suggested heuristics - 

should be go down and spin off dedicated drafts for OSPF and IS-IS to 
explicitly advertise the
transport IP address ?

>> 
>> for IS-IS traffic-engineering deployments:
>>  - if a /32 reported in any IP Reach TLVs (128,135, 235) matches  the
>>     TE-router ID TLV 134
>> 
>> for non IS-IS traffic-engineering deployments:
>> - if a /32 reported in any IP Reach TLVs (128,135, 235) matches  the
>> - IP interface address TLV 132 (and this is the only IP interface address 
>> TLV advertisement)
>> 
>> 
>> 
>> 
> 
> 
> 


_______________________________________________
rtgwg mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/rtgwg

Reply via email to