Re: [External] Request for information - Challenges in routing related to semantic addressing

[Geoff Huston <[e-mail]>]

> On 2 Apr 2021, at 7:15 pm, Luigi Iannone <[e-mail]> wrote:
> 
> Geoff, 
> 
>> On 1 Apr 2021, at 22:03, Geoff Huston <[e-mail]> wrote:
>> 
>> 
>> 
>>> On 15 Feb 2021, at 9:40 pm, King, Daniel <[e-mail]> wrote:
>>> 
>>> Hi AB, 
>>> 
>>> Thanks for scanning and commenting.                                                                                   
>>> 
>>> The I-D has bubbled for a while after seeing various discussions on contextual and semantic routing I-Ds, mostly clean-slate approaches. Although the initial literature review highlighted some common objectives across proposals and implementations, I struggled to find anything that defined "semantic routing" and goals that we could use/reference. I likely missed prior research that provides a useful definition. If not, then I agree that having the IRTF research community document a description for "semantic routing" and common goals, would be beneficial.
>>> 
>>> BR, Dan. 
>>> 
>> 
>> I take it you are referring to an approach being used in one of the largest IPv6 deployments: https://meetings.ripe.net/see3/files/Ian_Farrer-The_Terastream_Native_IPv6_Network_Architecture.pdf - Slides 15-17. If so, then I’d suggest this has already headed well beyond research into operational deployment!
>> 
> 
> Are you saying that the operator can give whatever semantic they wish to their addresses? We do not care?


well yes. For almost every value of “we”. As I recall, the motivating observation for these folk was that it was faster to place policies in a place that was processed by the router’s packet forwarding logic than it was to place these policies in a router’s filter list processing logic.


> 
> Even so, isn’t Terastream just a specific solution for a specific use case? 
> 


I dunno. The argument about preferring faster and cheaper tends to more than an isolated case.


> There is much more to be explored and discussed IMHO.
> 


That may be true, but my point was that this is not a matter “in the abstract” and one that is already in use. I was pointing out that there is one large network infrastructure operator has openly talked about the use of address structure in a way that this thread calls “semantic” addressing. There may be other similar use cases. Similarly, I'm aware of some (many?) ways to use the 64 bit interface identifier, some of which embed one-off tokens in the field in an extreme case of privacy addressing.


Geoff