(A-)Symmetric Internet Access

The Internet access specifications for residential cable-internet-access and ADSL-internet-access are always asymmetrical: the downloadspeed is larger (much larger) than the uploadspeed. Internet access over fiber shows higher speeds and quite often (but not always) the same assymetry where the download is much faster than the upload.

A lot of people wonder why. What is the driver for this difference? Technology? Costs? Marketing? Based upon a recent discussion between experts I have tried to summarize the geeky details. 

We have to keep in mind that Internet access was piggybacked upon networks designed for quite different purposes: cable for distributing TV-channels, copper network for telephony. So the engineers charged with creating Internet access had to work with what they got.

The key engineering issues involved in creating Internet access over these lines are attenuation, noise and cost of equipment. 

Attenuation is the weakening of signal strength as you get farther away from the source. If the strength of a signal starts to drop to the same level as the noise on the line you start to lose information carrying capacity (down to zero: no information). If you send information over a cable/telephone wire/fiber the signal strength weakens with the distance travelled. 

The rule of thumb is:  higher frequencies weaken earlier than low frequencies ( compare it to driving to a band playing on a festival ground: you hear the base player from far, the high notes only when you are close). 

The second rule-of-thumb is: the higher the frequency you can carry over a wire over a distance, the more information (bandwidth) you can send per second.

Telephone wires are difficult to work with: high frequencies weaken very fast with distance. Coaxial cable is much better (that’s why Siemens invented it : transatlantic telegraphs through normal wires was terrible) but anything over 1 Ghz weakens very fast. Fiber is way, way better (a million times better), you easily get 40-50 miles at extremely high frequencies before you have to repeat the signal. (Wireless behaves more or less the same and at the same time different: the signal weakens with distance but the sensitivity to anything inbetween you and the sending antenna is very dependent on the (carrier)frequency used.).

The noise you pick up is the second limitation. 

A normal copper wire acts as an antenna : while transmitting information over the wire, the wire acts as an antenna, sending at the same time to everything in the neighbourhood. 

It is also a giant receiving antenna, picking up noise and signals from neighbouring wires (crosstalk) and spikes (especially in old networks). This is why you run into problems if you get too many DSL subscribers in the same cable: your neighbours signal is noise to you. A coaxial cable is much better: it shields its transmissions more, and picks up less. But any endpoint or crack in the shielding introduces noise.

Fiber (again) has much less problems: it does not send anything, does not pick up anything. (But bad engineering, construction or maintenance can reduce its capabilities).

When DSL was engineerd for copper wires the design choice was to reserve a relatively low frequency band for the upstream link. The pro’s: less attenuation over distance so less sensitive for noise, less computing power needed at home (the required Digital Signal Processors where expensive at that point in time) making the gear cheaper. The con: limited bandwidth upstream but hey, who would ever need that? 

The general idea at that time was that users would passively look at websites and send some short email messages every now and then. In that frame of mind you need much more downstream bandwidth than upstream, but you need reliable upstream. Why? 

The TCP/IP protocol requires that the receiver sends an acknowledgement of received packets to the sender. Erratic or slow sending of ACK’s is a trigger for the sender to reduce the sending speed and re-send (presumably lost) packets of information. So for a fast download you need reliable upload, preferable in a ratio of 5 to 1 or better. (Try filling up your upload capacity in a normal ADSL line and you will experience a fast drop in download performance).

Downstream was positioned in a higher frequency band. The downstream bandwidth has been increased as DSP’s grew less expensive, but became less predictable as a result. It is dynamically adjusted to the maximum throughput possible, given the noise and attenuation. If you are unlucky (bad wire, noisy environment) the downstream advertised as 20 Mbps (max) is no more than 2 Mbps in reality (as I have experienced).

A lot of extremely clever engineering has been thrown at the problem to maximize the use any bit of untainted bandwidth/spectrum, but you cannot defy the laws of physics.

So VDSL was introduced: reduce the wire length to a couple of hundred meters so you can use higher frequencies (which would weaken too much otherwise) and gain bandwidth. The problem of not being able to predict the noise and crosstalk levels remains, so the predictability of the preformance is not good. Again the design choice in VDSL washow to split the available bandwidth. You can go for symmetric if you want to, but 40 Mbps down and 10 Mbps up advertises better than 25/25.

In Coaxial cables you have to allocate the available radio spectrum to downstream, upstream and TV channels. The upper limit is 1 Ghz or less, so there is a contest for space between Internet access and (HD)TV-channels. Again, very clever engineering squeezes a lot of bandwidth out of the spectrum at the expense of sensitivity to noise.

As the medium is shared by users (couple of hundred on one coax cable as a rule) you share the available bandwidth.

The standard upstream channel is limited in bandwidth and placed at the low end: again with a “content consumer” frame of mind and the objective to reduce the sensitivity for noise. Most hybrid fiber coax networks have not been designed for more upload capacity to begin with. Modyfiing the hardware is required as a start.You could also change the allocation of capacity in a spectrum (quite an effort but doable) or split up the coax network more so you end up with less shared users (messy, means digging and investing in hardware).

Again: physics and historical engineering choices have biased the access network towards asymmetrical bandwidth.

In fiber there is no such inherent bias. Symmetrical bandwidths are as easy as anything, most professional communication lines are symmetrical. Asymmetry is a design decision.

Even with GPON-fiber networks (USA, Japan) where the bandwidth is shared : symmetrical bandwidth requires a home “modem” designed for the task, but that is a choice you make. The GPON standard specificies an overall downstream bandwidth capacity which is twice the size of the upstream capacity. Again, a choice, not a necessity.

There might be a commercial reason not to offer symmetry. If you are a telco who has been serving high-end business customers with dedicated fiber lines, you have a problem explaining the difference in price between what they charge now and the consumer FttH offerings.

There is a reason why the business lines are/have been much more expensive (as a start: a dedicated dig is terribly expensive compared an FttH roll-out spreading all costs over many users) but it is a very difficult sell. An asymmetric offer makes the explaining a bit easier.