Portless networking and container communications

The shift of computing from hosting to cloud to serverless has a parallel in communications technology and products. “Portless” is the new “packet”.

Focus here: Wardley map showing the evolution from cloud to serverless computing


“Research is to see what everybody has seen and think what nobody has thought.” — Albert Szent-Györgyi

As part of my ongoing quest to be neither the smartest nor dumbest person in any room, I seek out experts to learn from, and those whom I can teach what I have learnt. Over time, I have discovered that true genius is often “obvious” in retrospect, since it involves uncovering an essential simplicity that others have overlooked.

I don’t often put new things into my “genius” category: quality attenuation network analytics and the ∆Q calculus would be one, the RINA network architecture is another. Wardley maps are the “genius” business equivalent of these technical concepts: an idea that makes one wonder “how did we get this far without it?”. If you can, and I strongly urge you to make the time, listen to Simon Wardley present his concept in this keynote presentation.

The best 23 minutes you are likely to spend on YouTube this week.

Wardley maps in brief

Rather than duplicate this information, let me just capture the essential points for those in a hurry. Firstly, this is a true business strategy map: it has all of the properties of a map, being visual and dynamic, with oneself being contextually located in that landscape. A SWOT chart, or Porter’s five forces, don’t have this true map property.

Secondly, it is anchored in the end user value, with the “value chain” visibility axis. The Apple logo on your iPhone is very visible, and the Apple experience and brand is figural. You yourself chose iOS over Android (or not). The brand of the manufacturer of the cardboard delivery packaging is not visible to you, and you didn’t choose them yourself. The brakes in a car are part of the “visible” user experience, but most car brakes are sourced from anonymous multinational conglomerates; but not always.

The usually invisible, made visible.

Thirdly, on the horizontal axis we have the evolution of the elements of the value system from their technical genesis to complete commoditisation. These utility systems then become the genesis inputs into new higher-order technical systems. Competitive advantage comes from the savvy integration of existing value elements, and differentiation from the successful inclusion of new “closer to genesis” subsystems.

When I worked at Sprint in the early 2000s, every device needed performance integration testing before use. BYO plans depend on new commodity inputs.

Finally, and in my view most critically, these maps also treat the management system as a first-class object in strategic planning. In Simon’s keynote, he considers traditional enterprise IT systems, with separate development and operations phases. This required the creation of capacity planning to bridge the two. Then as we moved to cloud, we aligned the incentives of these silos with DevOps. The operational management theory (and its toolchain) is a “technology” just as much as any other input into this strategic “system of systems”.

I am doing Simon’s work a disservice here for the sake of brevity. To learn more, you can read his free book, take a paid online course, or just build a map for free.

Cloud servers and network ports

Cloud is essentially the old idea of mainframe and minicomputer timesharing reinvented for the packet networked era. Virtual hosts offer a complete server, with its operating system as baggage.

This model is already rapidly approaching utility status, through services like Amazon AWS and Microsoft Azure. Whilst these may not yet be fully interchangeable and interoperable, in the way that different gas-fired or oil-fired power generation plants can be, we are not so far away. As such, cloud is already becoming the “new legacy IT estate”.

Meanwhile, all current cloud access network architectures and telecoms business systems aim to sell you a conceptual port in a switch, tied to a circuit of some kind. This can be a physical port, or a logical timed subdivision of a port (as with TDM or Ethernet). For any telecoms strategy problem, it’s historically been “The answer is ‘make and sell another port’. Now, what is your question?”.

After all, it’s baked into the protocols: IP explicitly addresses a port, with a port number. SS7 is equally tied into a port-to-port model. MPLS might was well have been “multi-port” as “multi-protocol”; it’s all locked into the same conceptual model. Every single mainstream telecoms initiative today — 5G, NFV, SDN, SD-WAN and more — exists in service of extending this port paradigm.

Serverless, meet portless

As Simon himself points out in the context of computing, the next wave is serverless, where we instead sell individual compute functions — “lambdas” in the jargon — which users call remotely. These are then formed into service meshes and distributed applications. Blockchain is the distributed system meme of the week (at least until everyone discovers a “flat and global” truth and trust model is a bad idea, just like it is for Internet Protocol).

Now I can deliver the punchline: the next wave in telecommunications is portless networking. This offers the complementary communications capability for serverless computing. It’s an inevitability, because the computing dog wags the telecoms tail, has done for decades, and there is no prospect of this changing any time soon. As such, it is a revolution as big as commercial packet networks arriving in the 1990s.

If you want a metaphor (and I insist you get one anyway), then think of packet networking as having been like inventing lightweight and strong corrugated cardboard boxes. These can replace unprotective thin sacks, and hard-to-move heavy wooden chests. It’s an improvement in how data transport is made affordable for all types of application.

Today’s packet networking model is still “break bulk”, schlepping all kinds of different “cardboard boxed” application traffic types together in the hold of “steam packet ships”. SDN is optimising the ship’s steam engine; NFV is upgrading the warehouse systems; but it’s still basically “burn coal and full bandwidth ahead” to the next port.

In a portless model, the service provider is (explicitly or implicitly) on the hook for delivery not to the cloud “harbour”, but to the final end customer. No leaving paper packets out on the quay to be pecked by seagulls, or lost in transit as the white delivery van weaves its way around the neighbourhood! Note that it’s not necessarily an end-to-end delivery contract, but a series of synchronised contracts the deliver the desired end-to-end result.

Furthermore, the contract is not for some rough aggregate performance of all packets, but for making the recipients of individual deliveries happy.  Needless to say, that’s a significantly higher bar to get over than “best effort” delivery of a domestic broadband line. If you think network bandwidth slicing is hard, this network latency dicing is a whole new game. We now need to manage the network at far finer granularities, and way shorter timescales. That’s why we need new maths like ∆Q, and new “high frequency trading” mechanisms, so as to assure performance for individual remote procedure calls.

“Portless” networks will need also “container communications” to aggregate demand into viable contracted “lambda cargo loads”. Physical 20 and 40 foot containers are highly regular, stack on top of one another, and they cannot contain other containers. In contrast, virtual telecommunications containers are irregular, “compose” (instead of stacking), and can be hierarchical. This is the basis of a new multi-modal data logistics industry, aligned to the scope-centric reality of distributed computing.

Portless networking is why we need new architectures like RINA, where the network addresses individual computations (which can be sub-containers), and not switch ports. The RINA framework can “wrapper” legacy port-centric transport protocols, doing to MPLS what MPLS did to SDH and ATM. RINA then acts as the “hypervisor container” for “portless” networking, abstracting away the physical location of the port from its logical function. There’s no traumatic “rip and replace”, but rather an evolution of the old into the new. The real barrier is, as always, in the billing and operational support.

The next iteration of the great telecoms game

There is a strategic pattern here that’s played out many times in the tech and telecoms world. For each “grand cycle” (punched cards, mainframes, minicomputers, PCs, smartphones, wearables, etc.) there are naysayers who point out the vast inertia of the present system, and tell you that the status quo is here to stay. But they are wrong every single time, if you’re reasonably patient.

As Wardley maps tell us, everything moves towards commodity status over time. The new wave finds its entry points and always disrupts the old. Better management methods arrive, displacing legacy working practises. For telecoms, the MBAs will soon find that you can successfully apply lean and six sigma quality techniques to portless (and packet) networks. CFOs will insist on the elimination of waste, and seek the resulting cost saving, especially once their competitors have seized the opportunity.

The strategic consequence of portless networking cannot be overstated, in the same way packet networking was previously underestimated. For example, it may trigger the ultimate demise of the present cellular ecosystem. 5G is trying to solve “portless” industrial problems with (in telecoms terms) stone and iron age “port” tools such as Internet Protocol and GPRS tunnelling. It’s just not going to happen.

The new “hyperinflation” of end points is based on different models, like WiFi, Bluetooth and IoT protocols. If portless networks work to extend these models, which have less incumbent friction, then 5G will never deliver anything beyond a capacity increase for traditional port-centric mobile broadband. That’s potentially a lot of stranded capital assets and unhappy investors.

Never, ever “buy back up the curve”. Another 6 minutes extremely well spent.

Serverless computing makes portless networking an absolute strategic certainty; “no-IP” is someday going to be the new “all-IP”. It’s already de facto the reality inside data centres, which run custom protocols that bear little relationship to traditional TCP/IP. The only questions are timing and tactics. Serious companies like 128 Technology are already rethinking the incumbent WAN model, offering a potential migration path. This is not a pie-in-the-sky theory.

Are you “investing down the curve” into the portless growth inevitability, or “up the curve” into the legacy ported model? Maybe you’d like a useful strategic map to help you decide what to do next? Just hit “reply”… but I can’t promise any genius, just common sense and experience.

For the latest fresh thinking on telecommunications, please sign up for the free Geddes newsletter.