Semantic Networking is the idea of making forwarding decisions in a packet network based on additional information carried in the packet (other than just the destination address) and informed by routing processes that consider network state and capabilities beyond the least cost shortest path.
To some extent, all packet networking is Semantic Networking. That is, network state information is distributed, routing algorithms determine the best paths, the forwarding components are programmed, and packets are forwarded based on the information they carry in their headers. Semantic Networking develops this by enhancing the information in the system.
Numerous approaches to Semantic Networking have been tried over the years. Typically, these mechanisms have been aimed at resolving specific problems applied to individual technologies, or have been intended for use in walled gardens or limited domains. Some of these techniques have seen successful (if limited) adoption and deployment (for example, Differentiated Services), some are on the cusp of fulfilling their potential (for example, Segment Routing), and some were just research projects or blind alleys. In general, each idea has been targeted at a specific function or behaviour and has been developed in isolation from other solutions making for a complex arrangement of stovepipes in a deployment, and a mess in router codebases.
Routing Area Working Group interim meeting
On 21 June 2022, I coordinated a virtual interim meeting of the IETF’s Routing Area Working Group (RTGWG) to discuss Semantic Networking. More than 60 people joined the Meetecho call to watch presentations and exchange their ideas.
A challenge with this sort of meeting is to bridge the gap between those who are familiar with the topic and those who know a lot about the routing system but haven’t done the homework and might have some basic questions such as:
- ‘Isn’t this just recycling New IP?’ (No, while some aspects of New IP might be classed as Semantic Routing, the objectives are very different)
- ‘Aren’t there privacy implications like in the early versions of Application-aware Networking (APN)?’ (Privacy needs to be a foundation of all new work in the Internet, and identifying applications or users in the headers of data packets is anathema).
So, it was with some relief that the meeting opened with words from Daniel King of Lancaster University to provide a quick refresher on the concepts and an overview of the problems that are motivating the discussion. There is a coming together of new technologies (such as fibre to the premises, 5G, and Low Earth Orbit satellites) with new and possibly futuristic applications (including holographic conferencing, collaborative video streaming, home market AR and VR equipment, massively multi-player gaming, and telesurgery). These applications each demand very specific traffic delivery behaviours from the networks that serve them. Some may need very high bandwidth, others may want ultra-low latency, bounded jitter, low loss, high security, and super resilience.
I like to get a view from the operator in these discussions, and so it was great to hear BT’s Phil Eardley talk about how applications evolve. Most importantly, he observed how Quality of Service (QoS) buys an operator time by enabling them to fulfil the requirements of a subset of applications with slightly less bandwidth but doesn’t create capacity in the network. This leads, inevitably, to the big question of the commercial realities of building and deploying networks that offer ‘advanced services’.
At this point in the meeting, I gave myself time to get on my hobby horse. I am really worried that the proponents of many of the modifications to routing and forwarding are too focused on their niche problem space and are not paying sufficient attention to the impact they may have on the broader Internet ecosystem. I’ve been talking about this concern for a while now (listen to the APNIC podcast) and I am collecting the concerns, issues, and best practices in an Internet Draft.
For many, of course, this is just obviously good engineering, but I think we need to hold our feet to the fire to ensure that proposals are properly thought through, designed, and tested. Part of that work should, in my opinion, extend to the higher-level architectural considerations raised by Semantic Networking in what Dirk Trossen calls ‘Routing Beyond Reachability’.
The discussion in the meeting and the chat room was lively, occasionally focused on the question at hand, and sometimes provided a good background on the theory of addressing. It brought up several points that were valuable to me, including:
- What is QoS? Isn’t it just deciding whose packets to drop? Is Semantic Networking ‘QoS on steroids’ or does it go further by dynamically choosing other than the shortest paths?
- What are the commercial realities of deploying advanced connectivity services for the customer? Is there a demand, how would the use of the services be policed, and are there regulatory concerns?
- What are the broader architectural considerations? What is it that is identified by an identifier and what is the namespace?
All of the materials, minutes, and a recording of the meeting are available via the RTGWG website.
What happens now?
What the IETF does now is up for grabs. The Routing Area Directors have steered further discussions to the Routing Discussion mailing list from where it could be picked up as a topic for more research and investigation, or it could be buried as another of those wild ideas that are more heat than light.
Adrian Farrel is a free-lance worrier about the Internet routing system.
The views expressed by the authors of this blog are their own and do not necessarily reflect the views of APNIC. Please note a Code of Conduct applies to this blog.