“Inside every router on the Internet, there is processor, which has some little-known and most magical properties. As each layer of atoms is laid down upon the silicon substrate, a Tibetan monk nearby prays that it will serve its new network masters well. Then, as the assembled equipment is packaged for customer delivery, traces of purest happiness, extracted from the tears of laughing human babies, are infused into the atmosphere of the packaging.”
Of course, all the above is total and utter tosh. The odd thing is, the broadband technical and policy community is unconsciously and unintentionally acting is if it were true and common sense. Indeed, it is a twisted parody of some nonsensical philosophical beliefs upon which we are attempting to build a digital society.
The semantics of network engineering
In our network economics and performance engineering workshops, I and my colleagues present the following chart. It is used as a teaching tool to discuss “what is success?” in data networking. The chart helps us think about the distributed computing experiences that we deliver to millions of users, and the billions of packets that enable the performance and availability of those experiences.
There are three basic components to “success” in networking (and every other domain of engineering endeavour):
- Intentional semantics: What do we want the system to do? You need to have a sense of the possible, and want the right thing.
- Denotational semantics: What can you ask the system to do? There needs to be a way of asking for what you want, and you need to ask for it.
- Operational semantics: What does the system do? There must be mechanisms for the system to do what you asked for, and then it must do it.
A looser way of say the same thing is that we have:
- intent (“entertain me”),
- purpose (“play this movie on Netflix by placing a load on the network”), and
- fitness (“damn, it’s buffering and too low a resolution”).
“Think what you want, say what you want, do what you say.” If you get these three things right (enough) and aligned (enough) then you are a winner: fit-for-purpose outcomes that reflect your intent.
An example: baking a birthday cake
Let’s work through a practical example. Imagine you are putting on a birthday party for your young child.
- Intentional semantics: Make a birthday cake and delight your child and her friends.
- Denotational semantics: Write a shopping list for your spouse, and you point him to the page in the bakery book with a great recipe, and tell him when you need the cake by.
- Operational semantics: Your spouse goes on a shopping trip, and he buys some extra medium eggs as the large ones have run out in the store, then stirs all the ingredients together, and bakes them.
You now have happy children (and no cake left). That is success.
In you had chosen the wrong intention (“delight children with personal copy of tax law book”), it would not have mattered how well you executed the rest. If you had chosen a recipe that could not be executed (“fresh Hawaiian white pineapple cake”), you are equally in trouble.
Furthermore, you have to be ready to adapt to possible failures, like swapping medium eggs for the out-of-stock large ones in order to get to the same outcome. If there had been no eggs in stock at all, buying a ready-made cake would have fulfilled the intentional semantics.
A ‘robot shopping parent’ following the literal denotational semantics would have failed in this task: it would explicitly have had to be directed to ‘buy ready-made cake’ as a ‘plan B’. That’s why online grocery shopping sites have an ‘accept substitutes?’ check box and comments section, to denotationally capture more of your intention.
Broadband: what do we want, ask for and do?
Moving back to the world of broadband, the failure to do what users intended or asked for manifests itself as (unwanted) performance hazards and (excessive) risk to the user. Hence we are likely to be interested in these forms of failure, if we are to comprehend the nature of success. Indeed, the language of broadband success is ultimately the paradoxical expression of what kinds of failure (i.e. performance hazards) that we are prepared to tolerate!
In the case of broadband, and specifically that of Internet access, these three semantic categories work as follows:
- Intentional semantics: Users typically demand and expect ‘everything will work for everyone all the time’. They have a sense of being annoyed when that fails to happen (e.g. your work Skype calls fails when a large software download starts on another home PC), so they do have preferences for the order in which things fail when not everything can be done at once.
- Denotational semantics: There is no direct means to express to the network the performance requirements of an application, or the order of failure. TCP/IP and media codecs attempt to competitively grab whatever resources they can from each other, as if it is their destiny and entitlement to achieve their individual application’s goals.
- Operational semantics: Mostly good things happen – outcomes are fit-for-purpose – but bad things can and do happen. Which things turn bad and when is largely non-deterministic (i.e. ‘random’ would be a serious improvement; the order of failure is utterly capricious).
The fallacy of benevolent teleology
Now we come upon two basic fallacies which pervade broadband networking, particularly as relates to the delivery of Internet access.
The first is that because we experience lots of successes at low network loads, we incorrectly infer this correlation to be causation: that the fit-for-purpose application outcomes were a result of the system understanding and acting on our intentions. This implies there is some kind of deus in machina which is divining our intention and acting upon it. The underlying false belief is that there is a teleology to networks, which means they have an intrinsic directedness, in this case towards meeting our human needs.
The second error is to believe that this deus is a benevolent one, simply because we have historically experienced relatively few major disasters. However, you can’t make valid statements about the likelihood of catastrophic (‘black swan’) tail risk by sampling data about the past. That’s a misuse of probability theory, one that Nassim Taleb has written about extensively. You might be reading this over a network with a wicked diabolus in machina instead, which could crumble at any moment!
The reality is that packet networks are not teleological. They are just bunches of multiplexers that schedule resources for distributed computations. There are no network gods divining our intentions, or devils working to frustrate them.
Instead, the long-term success of broadband requires us to be wise to the true causal relationships that exist between intent, purpose and fitness. Furthermore, we require appropriate incentives and controls to align them: we must think more clearly what we want to succeed, and what failure we can tolerate; say unambiguously what resources are needed for success, and what demand can yield in overload; and then ensure there are network mechanisms to deliver upon this.
Anything less leaves you at the mercy of the bandwidth gods, and their benevolent response to your wishful blessings and weeping babies.
To keep up to date with the latest fresh thinking on telecommunication, please sign up for the Geddes newsletter
Read It Now
