The unconscious and near-universal belief is that packet networks are a telecoms service, and one that constructs an ‘additive’ resource called ‘bandwidth’. This is demonstrably technically false. They deliver distributed computing services, as they calculate how to divide up an underlying telecoms transmission resource.

The ubiquitous error is a failure to recognise that the hardware platform has already been devoured by the software industry. Computer science is now the core of communications, not the underlying transmission mechanisms.

The result of this error is an inappropriate focus on packet forwarding ahead of packet scheduling. This, in turn, causes much of the latent value of that physical transmission resource to be lost. The potential returns from the capital assets of the telecoms industry are far higher than what is presently being delivered.

* * *

The historical model of telecoms

The basic model of telecoms is rooted in building physical plant: “holes and poles” that traverse the landscape. It involves getting out there in difficult terrain and inclement weather, and doing archetypally butch and manly things.

This physical plant is then festooned with technical mechanisms, from Morse code keys to pulsed lasers. These have generally been produced by the endeavours of yet more men (with some notable exceptions), who get praised for their masculine inventiveness. Grand prizes and accolades have been handed out by the hundreds to people who could capture and control the power of electromagnetism.

The operational role of dividing up the transmission resource, such as with switchboard operators, was seen as rather girly, and was dominated by women when done physically. Even today, incoming demand for conversations is given mainly to female-dominated call centre and receptionist work, which is typically low-paid.

The final act of consuming the resource required humans to invest their time in direct proportion to the number of bits demanded, be it telegraphy or telephony.

The packet data era arrives

We soon moved from humans listening to the telephone earpiece to computers, who have far more patience and can be made in factories. Whereas the network previously only had to support your chatty auntie, it now had to support hordes of replicant processors, capable of gluttonously devouring unlimited data.

The bursty nature of computer data meant the circuit resource sharing model had to change in order to be sustainable. When we wrapped circuits with packets, we moved from fixed time slots to variable delay (and the potential for packet loss). In turn, this created variability to be managed in a distributed system at short timescales, where there had previously been none.

Because nobody really wants to have their information lost or delayed, the values and the rewards remained as they were. Customers seemed to be happiest when you forwarded all packets as fast as possible. The internal rewards were all configured to give social status and financial benefit to anyone who manufactured more packet “bandwidth”.

The work ethic run amok

An accident of timing and transition meant that this belief their job to be done was “bandwidth” was reinforced. The arrival of broadband coincided with great innovations in applications, notably the Web, which created a surge in demand as part of a virtuous cycle.

It also coincided with the arrival of fibre optics and resource sharing using light frequencies, providing a corresponding leap in supply. This flood of capacity meant it was temporarily possible to use idleness to manage quality, masking the underlying protocol scaling problems.

As a consequence, there was a near-universal choice of “work-conserving” queues, which forward packets immediately when the onward transmission goes idle. This maximises the “work in progress” of the network, ingesting as much demand as possible, in the hope of consequently constructing as much “bandwidth” supply as is feasible.

As far as resource division by packets was concerned, basic first-in first-out (FIFO) queues were selected by default, since they seemed to do the job. Little professional reward was on offer for fiddling with the scheduling of the data. Why would you give this your attention when public virtue was seen as supplying “fat pipes” instead of “lean flows”?

If you said anything to the contrary you were a traitor to the Internet cause: a Big Telco shill, pushing some awful trope about a world of finite resources, so as to justify rentier exploitation

The category error

Yet here is the problem at the core of the telecoms industry today. Packet data does not perform an additive or constructivist function. It merely takes whatever physical transmission resource is there, fixed or wireless, and divides it up between competing uses and users.

Language is important, and in retrospect we might have been better off if we have not called it “packet data”. If instead we had named it “Statistical Division Multiplexing”, it would echo the time division multiplexing of the transmission. We might then have retained the clue to its purpose: dividing up the finite resource.

This, in turn, would have naturally provoked a helpful question: what actually isthe resource? For the resource is not merely the next link, but is the complete system of transmission. Networks are trees of multiplexers and demultiplexers. What we care about is the overall resource of the system, not the individual links.

The inevitability of mathematics

The unhelpful belief that telecoms is an outgrowth of the construction industry has real consequences for both the user experience and investor economics. We are in the “third epoch” of telecoms, where the computer science of “dividing up” increasingly dominates the concrete-and-capex of “building up”.

For instance, here is the broadband experience being delivered to the house I am staying at in Dorset. It is incontrovertibly dominated by load-related factors, not “speed” or bandwidth. This is simply an inevitable by-product of mathematics; you cannot escape this ineluctable reality.

Conversely, here is the level of packet contention at my home near London in the middle of the night. Over-provisioning results in fabulous off-peak experiences, but absolutely appalling and insane economics.

Over-provisioning networks is a way of avoiding having to confront the computer science issues. We madly manufacture ridiculous amounts of bandwidth because we can’t figure out how to schedule packets and engineer performance.

The cheese has moved. Will you?

Today’s networks are designed to locally optimise link usage, whilst globally pessimising the user experience and cost structure. Yet maximising work-in-progress is not a good idea, as the whole history of ‘lean’ tells us. “Fat pipes” are a fat load of good if you end up in a fight with fundamental constraints.

The centre of gravity of telecoms has already shifted from hardware to software. Software-defined networking does the “low-frequency trading” of resources to make “bandwidth” more “plastic”. Next up is “high-frequency trading” of the resources to deliver a new “lean quality” flow model.

This is a basic shift in emphasis from the “build up” to the “divide up”. It makes computer science figural, not transmission and trunks. However, the beliefs and behaviours of this industry have yet to reflect this inevitability. Many of our current pains reflect the gluttony of unhealthy fat pipes where we ought to be fasting for lean flows.

* * *

Epilogue: A tentative theory of gender values

I have a hunch that there are deep cultural and psychological issues that are keeping telecoms trapped in a hardware mindset when we live in a software world.

My very lightly-held opinion is that telecoms is unconsciously locked into a masculine paradigm. Packet scheduling is the modern equivalent of the switchboard operator. Admitting that the “girly” has overtaken the “butch” upsets the core social and power structures of the industry.

This masculine “dominator” vs feminine “nurturer” is playing out on an epic scale across the whole of the technology industry and society. In the case of telecoms, the present model may never transition from masculine to feminine. Instead, it may take a new generation, with a sensual model for the sensor world, to displace and replace it.