Re: [arch-d] [icnrg] recasting address semantics

[Adrian Farrel <[e-mail]>]

This thread seems to have wandered around a bit: arch-discuss, ICNRG, RTGWG.
And yet, we're told that the home for discussing "semantic routing" is COINRG.

Anyway, since I coined the term "semantic routing" in this incarnation, perhaps I should comment?

I'm a little disturbed by the engineering interpretation of "semantic routing" like it has been proposed as a solution for a specific problem (or worse, a solution in search of a problem). Far from it! This work originates from observing that there are already very many schemes proposed, researched, and deployed that enable routing and forwarding based on more than the destination address, and more than the classic 5-tuple. We started to collect these schemes in draft-king-irtf-semantic-routing-survey although we got a little dispirited when we found out exactly how many such schemes there are.

Our next observation was that most of these approaches assumed deployment in some form of "limited domain". This limitation ranged from regular routing domains (areas/levels or ASes) to specific technology or use case environments. This meant that many different schemes were being proposed/developed with no consideration of interoperability and no thought to general applicability.

Of course, we have a proud history of developing routing systems for specific environments or use cases. Likewise encapsulations. But many of these ways of enabling semantic routing are placing new information in existing IP header fields (addresses, Flow Label, Traffic Class, ...), introducing new header fields (mainly as Extension Headers), or proposing new encapsulations. I was most concerned about the "overloading" (or meaning change) of existing fields, because this seemed often to not consider what happens if a packet "escapes" the limited domain, when domains are merged, or when traffic has to transit multiple domains. So we started work on draft-king-irtf-challenges-in-routing as a way of capturing the things people should think about when proposing new semantic routing schemes. This rapidly grew into considerations for all new routing research as we have observed that routing research often experiments with very simplistic topologies and is rarely independently verifiable.

As we started to talk about these two drafts, people kept asking "Yes, but what do you mean by Semantic Routing?" So we produced draft-farrel-irtf-introduction-to-semantic-routing to describe the sort of schemes we are looking at. Sadly, many have assumed that this document is a proposal to build a specific solution rather than a scoping of the problem space. Of course, it would be good to move on from this introduction to see whether an abstract definition and architecture of "semantic routing" can be produced, and that might lead on to research into a more generic solution (which can be seen impinging on IETF discussions, such as APN).

Further discussions in the COIN context (especially with regard to seeing how Semantic Routing might be in scope for COIN) led us to consider network programmability and network compute. The first of these clearly splits into programming of devices, and programming of the whole network: draft-boucadair-irtf-sdn-and-semantic-routing starts to collect thoughts on this. The second consideration needs further thought, but topics might include how to install different routing or forwarding algorithms on network devices.

(Ideas of "Active Networking," with micro-programs embedded in the header fields of packets, are not in scope for *our* discussion of Semantic Routing.)

Why is this all relevant to this thread on "recasting address semantics"? Well, many existing Semantic Routing schemes rely on placing additional meaning on the IPv6 destination address. These approaches may be "safe" because they only play with the low-order bits of an address that are (probably) not used by existing routers for routing purposes. Others may be less easily integrated into existing routing systems with the result that we are looking at network-wide upgrades within out limited domains, significant risks resulting from leakage, and gateways needed at the domain borders.

And, finally, it was at the recent COIN interim that "all routing is semantic routing." To some extent that is true, but it would be better to say "we have always supplemented routing on the destination address with routing on other fields in order to achieve service differentiation." What we are seeing now, however, is all sorts of different niche requirements to provide enhanced service differentiation in the network (enhanced VPNs, network slicing, latency aware routing, energy-aware routing, ...) with no coordination, no big picture, and no consideration of the impact on the stability and utility of the integrated whole.

So, I think 
> The more I read about the “semantic routing” work the more pessimistic
> I am that it will lead anywhere useful.
Indicates a failure to understand what is being talked about.

Cheers,
Adrian

-----Original Message-----
From: Architecture-discuss <[e-mail]> On Behalf Of Dirk Trossen
Sent: 25 February 2022 08:35
To: Geoff Huston <[e-mail]>; David R. Oran <[e-mail]>
Cc: [e-mail]
Subject: Re: [arch-d] [icnrg] recasting address semantics

[SNIP - not to judge your observations negatively but merely agreeing with them at large] 

The strength of the innovation of the Internet for me lies in the deregulated nature of the activity. There is little to say “you can’t do that!” But the market is also a harsh judge, and concepts that are not economically viable are rapidly discarded, irrespective of any other merits they may have had. In my view what seems to work these days are evolutionary paths that do not attempt to redefine the underlying existing infrastructure, but, in gross terms, ignore it as far as possible!

> 
> The more I read about the “semantic routing” work the more pessimistic I am that it will lead anywhere useful. Perhaps you will prove me wrong in the end.

I also don’t have much optimism about the prospects for “semantic routing”, but I am pretty sure that what will determine the fate of this and similar efforts is simply whether it achieves a stable and viable niche in the market space that collectively we call “the Internet”. i.e. the determination of success or otherwise these days is ultimately based on the deregulated market.

[DOT] we are coming to the crux here, namely should activities on routing innovation happen or not and, also, how should the IETF support such activities. It is one aspect that ANY innovation is judged ultimately by the market and I cannot and do not disagree with it. But the question here is whether activities on routing (whether 'semantic routing' or others) should be (mainly/purely?) judged by current economic conditions or should be fostered in the spirit of what you call a 'deregulated nature of the activity', i.e. enabling the possibility to form those stable and viable niches in the market (which themselves may grow or disappear altogether). 


_______________________________________________
Architecture-discuss mailing list
[e-mail]
https://www.ietf.org/mailman/listinfo/architecture-discuss