“Philosophy is a battle against the bewitchment of our intelligence by means of our language.” —Ludwig Wittgenstein

The words we use to describe telecoms networks often contain hidden metaphors and meanings that lead us into wrong thinking. Here are three examples.

#1: “Best effort”

Why misleading? The word “best” implies both benevolent and optimal intentionality: the network is going to do the “right” thing for its users, and it will maximise the “rightness” in some way.

The word “effort” falsely implies networks do “work”, with the expectation that ISPs are going to “bust a gut” getting the service to “work” all the time for you. Indeed, they are going to use their best endeavours to put all the “effort” they can muster into the service.

What they actually mean? You will get the technical embodiment of “laissez faire”. What you can reasonably assume about the performance on offer is… absolutely nothing! There is no implied worst case.

Furthermore, the past cannot be relied on to be an indicator of the future. Your application may work some of the time, but could fail at any time. If it doesn’t work, then you have no redress.

Better alternative(s)? “You got what you got” or (more formally) “nondeterministic performance”. Note that this is qualitatively worse than “random”, since random means it has properties that should hold over time, whereas nondeteriministic means no such assumption can be made.

#2: Packet “Loss”

Why misleading? Losing physical things (like umbrellas or children) is seen as bad, whereas in packet networks “loss” is vital.

What they actually mean? There are two types of loss here: 1. Loss due to transmission corruption (which is sort-of-bad, but sort-of-inevitable, at least at some rate); and 2. The choiceful discard of a packet by a network element.

The truth is that discarding packets is a good thing, and integral to a system of two degrees of freedom (among load, loss and delay).

It’s a bit like the “pressure, volume, temperature” you learnt in physics classes at school. If we want low temperature, we need low pressure and/or high volume. If we avoid “loss” then we must either have unboundedly high delay, or an extremely constrained low load.

Typical “loss” sources are the network defending itself from delivering even worse performance characteristics. This loss is an inevitable consequence of the statistically shared nature of broadband. Such statistical sharing is the key property that makes broadband affordable. So no “loss” means no mass market data service!

Better alternative(s)? More neutral terms like packet “erasure” or “discard”.

#3: Network “speed” (and hence “fast” networks)

Why misleading? The speed of light is fixed, so networks aren’t getting “faster” like CPUs do. Indeed, there’s no “Moore’s Law” for networks.

The idea of “speed” conflates packet (de)serialisation time (i.e. how long it takes to “squirt” the packet over a link) with low contention (due to network idleness). It also conflates “speed of response” (delay) with “speed of download” (capacity).

An unfortunate consequence of the ubiquitous belief in network “speed” is that more “speed” is always better. This is not true.

What they really mean? When someone talks about a network being “fast” they are typically referring to the perceived responsiveness of applications that they are using. These application performance outcomes are the result of the interaction of network performance, protocols and application design.

Better alternative(s)? Use “capacity” for capacity, “timely” for low latency and loss, and “responsive” for good (interactive) application performance.

* * *

These three are not the only examples of misleading language in telecoms. For instance:

• “Packet” data implies physical thing, which leads to an unhelpful “beads on a string” model. It also implies that the packet is the important thing, when what is important is the information it encodes.
• Conceiving of a network as a “pipe” implies it contains a fluid, with inherent sequential and smooth delivery over the same route, whereas packets are bursty and can take many routes.
• Even the word “network” leads us to think of a “flat” tangle of concatenated interconnections, rather than a “stacked” hierarchical set of scopes of information copying.

It is only by becoming aware of the hidden meanings of our language can we truly let it acquire clarity and precision: “To handle a language skillfully is to practice a kind of evocative sorcery.” —Charles Baudelaire