Hi Dino,

> On 24 May 2023, at 19:45, Dino Farinacci <[email protected]> wrote:
> 
>> there are a few things to ponder:
>> 
>> - Looking at lispers.net the 254 value choice, it looks like a quick hack. 
> 
> I would refer to it as a convienent solution that doesn't violate the spec.
> 
>> - What about backward compatibility? If we allow overloading, there is no 
>> way to understand whether a value indicates a “true” priority or something 
>> else, different implementations may interpret the value in different ways 
>> with unpredictable results.
> 
> It always means a true value from an xTR point of view.

Understood. So the most accurate statement is that it "indicates the priority 
AND additional information, i.e. it is a RTR”.  


> 
>> - What about weight? In the lispers.net NAT traversal it is used as defined 
>> in the main specs, but this means that all RTR have the same priority all 
>> the time. And what if a future value will indicate not to use weight? Or use 
>> it in a different way?  
> 
> This solution does not violate the LISP spec on how ITRs/RTRs use priorities 
> and weights.

You are adding protocol actions not related to priority upon a specific 
priority value…. Without calling it a violation of the specs this is a not 
following the specs as they are for now. Hence the question what should be 
done. 


> 
>> - With the above we end up having RLOCs priorities that can be priority or 
>> something else. In this latter case weight can or cannot be meaningful (or 
>> even be something else altogether). Architecturally speaking it looks to me 
>> less clean. 
> 
> This is simply not true.

Yes, this is not true in the lispers.net <http://lispers.net/> case.
> 
> I think you are overstating the problem.

If we open the door to the use of priority value that are different from just 
priority values the question is quite relevant IMO.

> 
>> Now, let’s take one step back: the real question seems to be how to signal 
>> in the mapping system that an RLOC belongs to a RTR? 
> 
> You could do it with a distinguished-name AFI that is encoded with the RLOC 
> address.

Excellent. So we have another possibility beside LCAF.

> 
>> The answer to me is RFC 8060. Just use LCAF! The LCAF format has 16 reserved 
>> bits. One can be allocated to indicate whether the RLOC address belongs to 
>> an RTR.
> 
> That could be too specific to this use-case. What an ETR needs to know is if 
> it should tag RLOCs it gets back from the map-server in Info-Reply messages 
> with this bit.
> 
> It actually could not tag it at all since the map-servers know the addresses 
> of RTR RLOCs when they advertise them. But that means all map-servers need to 
> have the same info and there is configuration coordination required.

Don’t see this. The bit (or tag) is associate to the RLOC, the same way the 254 
value is.


> 
>> A side benefit of this choice would be that older implementations will just 
>> ignore the bit, hence taking no action, rather than interpreting the bit in 
>> a different way. Looks like a safer situation to me. You can even use a 
>> whole new type, so that an implementation either knows how to handle it or 
>> does nothing at all.
> 
> No that won't be the case. The way it works is that the RTR will give an 
> RLOC-set to the ITR. So the ITR doesn't know if an RLOC is an ETR RTR pETR, 
> etc. But the ETR side registeres the RTR RLOCs with 254. That is what is 
> implemented today. If the ETR DOES NOT do this the map-server could figure it 
> out on its own (as I mention above).

I was not referring to you implementation. I was referring to possible 
interoperability issues.

Ciao

L.
> 
> Dino
> 
>> 
>> Thoughts from the WG folks?
>> 
>> Ciao
>> 
>> L.
>> 
>> 
>>> 
>>> Note this is only how the map-server operates. So existing xTRs will get 
>>> back whatever the map-server decides. So if you are not an RTR (that must 
>>> be configured in the said map-server) you will get back an RTR RLOC that an 
>>> xTR will happily encapsulate to. That is, it works with existing xTRs that 
>>> don't know anything about NAT-traversal.
>>> 
>>> This implementation has interoperated with other implementations, but we 
>>> don't claim anything in the draft. And existing xTRs can *receive* packets 
>>> without following the control-plane procedures from the draft. We 
>>> demostrated this with OOR by doing gleaning on the RTR.
>>> 
>>> I have videos demostrating this for unicast and multicast and can send 
>>> pointers if people are interested.
>>> 
>>> Dino
>>> _______________________________________________
>>> lisp mailing list
>>> [email protected]
>>> https://www.ietf.org/mailman/listinfo/lisp
>> 
> 

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

Reply via email to