Re: Comments on draft-manral-ospf-router-lsa-unknown-type-02

[Alex Zinin <zinin at PSG.COM>]

Acee, et al-

  After thinking more about this:

  We all seem to agree new link types explicitly affecting route calculation
  will need to ensure everyone in the area understands them. The remaining
  question is whether other things (those that we believe shouldn't affect
  normal route calculation) will trigger an unwanted reaction or not. The crux
  of the matter here is two-fold:

    - redefining the "unknown link-type" situation from an exception with
      unspecified processing to a normal situation with a well-defined behavior

    - whether such redefinition will be effective 'ex post facto', given
      current protocol deployments, and what are the worst-case scenarios

  One of the questions here is whether we sure there are no networks deployed
  out there that will react to an unknown link-type (when a newer router gets
  its SW upgraded) by ignoring the whole router-LSA as John suggested, or
  screaming loudly.

  Leaving this in the capable hands of this WG now.

-- 
Alex
http://www.psg.com/~zinin

Saturday, March 19, 2005, 6:16:27 AM, Acee Lindem wrote:
> Hi Alex,
> I don't feel that strongly on whether or not we need to document the
> behavior when an unknown link type is encountered. However, I was
> content to ignore them and leave the backward compatibility issues
> to those proposing new link types. If there were compelling reasons to
> introduce a new link type we are now relegated to a solution where we
> need protocol mechanisms to assure that every router in the area understands
> it prior to it's usage and, depending on the link type, this may be more
> conservative than what is needed.

> Thanks,
> Acee

> Alex Zinin wrote:

>>Vishwas,
>>
>>I'll try one more time.
>>
>>  
>>
>>>If someone wants to write an implementation of OSPF, what does he/she do
>>>regarding unknown link types?
>>>    
>>>
>>
>>We discussed this earlier in this thread.
>>
>>  
>>
>>>We need to define the behavior else we
>>>will never get interoperability. I guess we both agree with the above.
>>>    
>>>
>>
>>We do NOT have an interoperability issue unless we start deliberately
>>introducing new link types without making sure everyone understands them.
>>
>>  
>>
>>>If I understood the discussion in the mailing list right, most
>>>implementations ignore the link and that is what we are trying to
>>>standardize.
>>>    
>>>
>>
>>A few years ago we had a discussion that touched on link records. John mentioned
>>that there are still implementations out there that will ignore the whole LSA
>>when the see an unknown link type. I don't think we can ignore this.
>>
>>  
>>
>>>The suggestion you make will not interoperate with the
>>>installed base at all while a MUST ignore solution will with most of
>>>them.
>>>    
>>>
>>
>>I don't understand what you mean by "will not interoperate".
>>
>>Alex
>>
>>  
>>