Hanging up on the Phone System

I was fortunate enough to be invited to attend The End of the Phone System conference at Wharton in Philadelphia last week as a respondent to one of the papers. To me personally there is significance in the event existing at all, and having the name it does: it’s an acknowledgement of what I started writing back in 2003 on my old Telepocalypse blog that over-the-top (OTT) voice and messaging providers would cause the phone system to collapse at some point. This is both a fixed and mobile phenomenon, although the dynamics are somewhat different in each case.In the interim my understanding of why and how has changed a lot. The dynamics are more complex than just “the Internet and OTT wins”. There is value in the telco way of doing things – simple, convenient, ubiquitous packaged services. The social ideals and ideas of the PSTN are valuable, and the laudable goals of common carriage and universal service are in danger of being lost. The problem is that continuing that valuable role requires new approaches, technologies, regulations and business models.

The system is corroding

There are a huge number of structural elements that go together to make the PSTN (including the derivative PSTN-alike IP voice systems). These are political, social, economic and technical in nature; indeed, the PSTN can be seen as a social compact rather than a phone network. Their interconnectedness means that the system is (by design) highly stable and rigid. The problem is that the failure mode is not via a visible warping, but rather from miscalculations in specification and lack of maintenance causing corrosion of the load-bearing elements.

Externally the system appears to be functioning, whereas internally it is weakening. Structural failure could then be widespread and relatively swift, following some sudden stress on the system that forces a loss of confidence in its future. It is not possible to engineer a controlled demolition by exploding regulatory lawyers and lobbyists in strategic places. Rather, collapse could easily be associated with a mass-casualty (real-world, not metaphorical) event. This is not going to be pleasant, and bad law is often made in haste following such events. (For example, consider spectrum regulation introduced following the Titanic sinking, or in the UK the famous case is the Dangerous Dogs Act in 1991 following a media-induced political panic.)

The PSTN does not exist in isolation

Mobile over-the-top VoIP is growing fast, and there are significant pains in delivering a working official telco solution for voice over LTE. This means that telco voice as a whole is ultimately under threat. It is quite possible that Skype on 4G will on average work better than the official VoLTE solution. In my colleague Dean Bubley’s words, Microsoft bought an $8bn call option on the whole mobile industry. The increasing tendency of telcos themselves to go “over the top” with voice and messaging, and extra-territorial, is a Prisoner’s Dilemma game where “defect” is the increasingly-chosen option. (Comcast launched an OTT VoIP product this week, and soon every telco will need one.) No amount of work by the ITU and GSMA will move the apostrophe back to make this into a collaborative and federated Prisoner’ Dilemma instead.

This process is strongest outside the US, but will have a spill-over inside the US at some point. There will be an effect on the PSTN if mobile shifts voice use away from traditional PSTN-like numbering or calling patterns. There are numerous scenarios where voice fragments and we see private voice islands emerge. We already see this with regulated corporate landlines becoming deregulated SIP trunking services. Continued global voice network interconnection is not a forgone conclusion. The balance between traditional settlement and new peering arrangements could shift quickly.

The business model context is not static

As voice becomes a software feature embedded in different communications services and media, I expect we will see the business model shift from ‘connect’ to ‘interact’ and ‘transact’. Rewards will accrue to those voice services with value-added B2C and B2B features that make everyday business more efficient, effective and secure. Think of a “freephone on steroids and amphetamines” – stronger and quicker at doing business (albeit without the lethal health penalties). This could be a feature of iMessage, Facebook, or new services that enter the space.

Thus the context for voice service may shift rapidly, and enable all kinds of interesting arbitrage of existing regulations. For example, what if IP-based network operators offered a “conditional call termination API”, such a location- or presence-triggered call routing. This might have both regulated deliver and unregulated information components. You don’t need to be a very smart lawyer to see how to game the termination system by shifting revenues between the different components.

The Internet is not a pipe

The assumption in the room was that the Internet is a “dumb pipe” which telcos wish to avoid. This fails on several counts. Firstly, the PSTN is damned dumb (as is a T1, etc.), but nobody worried about that as the margins were fat. Secondly, it is not a pipe, in that pipes have one degree of freedom (pressure, and thus flow); statistically-multiplexed networks have two (loss and delay).

This matters, since the Internet thus contains multiple multiplexed flows with differing loss and delay needs. These cannot be serviced well via traditional QoS priority mechanisms, since they conflate loss and delay rather than manage the two degrees of freedom independently. Furthermore, today’s Internet is unable to “fractionally distillate” these flows-qualities out, or (unlike the PSTN) provide any bounds on the loss and delay. The inability to match packet flows to their transport quality needs limits the utility of the Internet.

The Internet is not ready to replace the PSTN

The Internet is fundamentally incapable of supporting reliable mass voice service, and always will remain so, no matter how much capacity is added. Capacity does not solve contention between correlated packet flows, and indeed can make it worse in some situations. This is subtle: the bandwidth religion just assumes you can “overprovision everywhere” and contention issues go away. This disregards the real-world behaviour, as you can’t have “fat” going into “fatter” in every direction. When “fat” goes into “thin” you have contention, and you quickly find yourself back to where you started if you inject packets in the right-wrong order.

Whilst at the Freedom to Connect conference in DC this week there was a delicious moment of gestalt when Cory Doctorow was presenting via Skype from London on how managed, controlled and closed computing and communications systems are bad. The Skype call repeatedly broke up. The answer nobody was willing to propose was to place a call via a landline or mobile device which has been engineered to carry voice.

Furthermore, the obese pipes approach places an unbounded burden on your capex to solve all quality issues with additional quality. How much electrical power are you willing to use to add bandwidth to solve your contention issues? The Internet is like a smokestack steelworks belching CO2 and fumes with little regard to real-world issues of efficiency or sustainability.

The Internet’s failure modes are mismatched to voice

As a first-order approximation, the PSTN offers a single quality class, assured service, and step function on quantity (admitted or busy). The Internet offers no quality assurance, and no admission control. Whilst users will accept a degradation in voice quality (cf fixed vs mobile), they hate interruption as it breaks the cognitive illusion of speaking to another person, rather than a lump of sweaty plastic pressed to their ear. The Internet can, and does, readily enter into breakdown from laminar packet flow (cf bufferbloat) and the problem is intrinsic to the use of control loops like TCP and exceeding the predictable region of operation as the network saturates.

There is no solution without an architecture change; the math tells us so.

The “reliably unreliable” nature of Skype et al is what we may have to live with for some time, as things will get worse before they get better.

A change is gonna come

Many telcos and cablecos imagine they are playing the game cleverly, and positioning themselves to either capture the value of the PSTN (by moving to IP whilst collecting the termination revenues but avoiding the social costs), or avoiding the issue by offering Internet-only service. However, humans greatly value being able to see each other and talk to each other at a distance, and will always place extreme value on this ability at critical moments in life.

The irresistible force of the human desire to talk is about to hit the immobile object of the PSTN’s demise, and the falling rubble is going to squash a lot of carefully-constructed temples to money-making located in the vicinity. That may include a lot of telcos and cablecos who rely on the continuation of both circuit voice and broadband ISP business models.

Conversely, this opens up an enormous amount of opportunity, both for opportunists who wish to scavenge and recycle the wreckage, as well as those willing to farm the new technologies and business models of cloud-based voice and messaging services. The need to deliver socially and economically essential voice communications is not going to go away.

[Editor’s note: This article was originally published as a Future of Communications newsletter. You can subscribe at www.futureofcomms.com]