40

This has been bugging me for a while now. I did a couple wireless transmission courses back in my university days and it's a known fact that wireless transmission is hard. Partly because it's a shared channel, physical objects can be in the way, air provides quite a bit of loss, etc.

So why is wireless transmission catching up to Ethernet in terms of transmission speed?

We see MKBHD in this video going to some street with a phone and getting over 1Gbps. Surely a reserved and shielded Ethernet channel should give us 1Tbps right?

I feel we are just not trying enough on the Ethernet side of things. Is it the case?

auspicious99
  • 326
  • 5
  • 17
Francois
  • 567
  • 1
  • 4
  • 6
  • Comments are not for extended discussion; this conversation has been [moved to chat](https://chat.stackexchange.com/rooms/102249/discussion-on-question-by-francois-why-is-wired-ethernet-losing-its-speed-advant). Any conclusions reached should be edited back into the question and/or any answer(s). – Dave Tweed Dec 16 '19 at 13:18
  • Just to note, from a theoretical standpoint wireless *should* have higher bandwidth than wired. Imagine if you can beamform point to point wireless signals between transmitters and receivers, then multiply by how many wireless paths you can have, and compare that to how many cables you are willing and able to run. – Michael Dec 16 '19 at 17:38

9 Answers9

64

Wired ethernet is not losing its advantage. There are standards out there for 10Gig Ethernet (802.3ae), 40Gig Ethernet (802.3ba), and 25Gig Ethernet (802.3bq). These are primarily backbone/backhaul standards.

It's not an apples-to-apples comparison to compare 20-year-old (1 Gb Ethernet) or 25-year-old (100 Mb/s "Fast Ethernet") technology to the latest in wireless just because that's what's available in the consumer market.

auspicious99
  • 326
  • 5
  • 17
BrianB
  • 966
  • 4
  • 9
  • 37
    In fact, 10G Ethernet IS readily available in the consumer market. – Dave Tweed Dec 13 '19 at 14:55
  • Sure - you can buy 10G equipment if you're willing to pay a lot of money. But what are the use cases for the *average* home consumer to need anything more the 1GE connection to a single piece of equipment? – BrianB Dec 13 '19 at 14:58
  • 2
    @BrianB A 10gbit SFP+ module costs about the same or a little less than an M2 802.11ax card. A 40gbit QSPF+ module cost $5-10 more. Switches are more expensive (since they have to be able to process incredible amounts of bandwidth), but WiFi cards are not particularly cheap compared to 10/40Gbit transceivers. – user1850479 Dec 13 '19 at 15:00
  • See [Linus Tech Tips](https://www.youtube.com/watch?v=b4Wp4SzdNj4). But that isn't the point of the question. – Dave Tweed Dec 13 '19 at 15:03
  • 3
    Yes but let say from year 2005 the average person wireless transmission speed went from 11Mbps to 1000Mbps+ (x100), in the same time range, Ethernet speed for average person went from 100Mbps to well 1Gbps (x10). See my point? – Francois Dec 13 '19 at 15:50
  • 23
    @Francois no, you don't have a point. You have a postulation "there should be a fixed ratio between wireless and wired speed", and that's not based on anything! – Marcus Müller Dec 13 '19 at 16:00
  • 11
    Ethernet standards exist for 200 and 400 Gb/s also. IEEE 802.3-2018 clauses 116-124. (802.3ae, 802.3ba, etc., were proposals and drafts, not official standards. These proposals are now codfied in various clauses of 802.3-2018) – The Photon Dec 13 '19 at 17:09
  • 18
    @francois tell me, have people been more interested in faster wireless or faster wired? Typical consumers want faster wireless, and that is where commercial business has been focusing for quite some time now. For most consumer needs, ethernet is a pain, and gigabit is *plenty* fast. – akohlsmith Dec 13 '19 at 17:14
  • 1
    I would question what the average person would need 10Gbps(x100) for. Development tends to follow demand. Wireless had to play catch up to reach the demand, Ethernet reached it a long time ago. – William Dec 14 '19 at 10:34
  • 4
    @akohlsmith I'd argue that Wireless is the actual pain - for all of us that are assumed to know how computer networking works and that get asked "why doesn't it work? Why is it so slow?! Fix it please!". The advantage that wired has independent of potential max speed, is its stableness. No interference no configuration problems. It just works - unless you physically break the cable. But you are totally right in your assessment what consumers at the moment mostly ask for. – Frank Hopkins Dec 14 '19 at 20:11
  • 2
    @FrankHopkins WiFi has the advantage of being universally a WiFe approved standard. Whereas ethernet compliance to WiFe standards can vary drastically between locations. – Aron Dec 16 '19 at 01:58
  • 1
    2.5, 5, 10, 25, 40, 50, 100, ... are all available on the market now for wired. Commercial access points are starting to use 2.5 and 5.0 to handle the number of radios in the unit. Servers are likely to move from 10 to 25 in the near future. 2.5 and 5 are using the existing cat 6 pairs. 25 and 50 are able to use the same copper that 10 uses (10 has been around for a long while now). 100 was using 10 pairs of 10 but 100 is moving to 4 pairs of 25 using qsfp28 instead of cxp. If you use copper, fiber is of course available but costly for these speeds. – old_timer Dec 16 '19 at 12:33
  • consumers will probably start getting desktop/gamer machines with 2.5 along with switches in the near future as the 2.5 gains traction. Lan party folks have had access to 10 for a while now although probably dont use it due to the cost even with direct attach copper (DAC) vs fiber. But even small lets say medium businesses that have an IT department are using or wanting 10 between their file system and servers or at least the backup system. 10/25G sfp+ boards have been out for a while so they can switch to those and get 25 when both sides are ready. – old_timer Dec 16 '19 at 12:37
  • 1
    @FrankHopkins but the sensible approach to "wireless is, at times, much faster than it needs to be, but the problem is it's not reliably fast enough" is making sure we raise the guaranteed availability, e.g. by more access points indoors, since it'll be hard to sell iPads with ethernet ports. – Marcus Müller Dec 16 '19 at 15:56
  • @MarcusMüller from a marketing perspective, sure, from a personal perspective: hell no, even more access points to configure and make sure they don't interfere. Btw. I'm fine with a wireless router for tablets and smarties at home, as that's the natural solution for them, and any "it doesn't work (fast enough)" can be answered by 'use the desktop / laptop with a cable then'. I'm not fine with wireless to bridge the rooms between the main router and whereever the desktop/laptop is mainly used, but recommend ethernet or will not do any fixing of "speed" issues etc. – Frank Hopkins Dec 16 '19 at 19:48
  • 1
    @FrankHopkins exactly, marketing! modern mesh solution (whilst still expensive) seem to be really easy (press button on access point or nearest mesh node, press button on new mesh node / repeater, done) – Marcus Müller Dec 16 '19 at 20:37
  • @old_timer in fact my last MB came with 2.5ghz ethernet. – barbecue Dec 16 '19 at 21:43
27

We see now the MKBHD guy going to some street with a phone and getting over 1Gbps. Surly a reserved and shielded Ethernet channel should give us 1Tbps right?

nope, why should it? Data rate is a function of SNR and bandwidth. (Shannon Capacity: you can only transport a limited amount of bits per second over a medium with limited bandwidth and limited signal fidelity; both applies to twisted pair; that's impossible to circumvent through research; it's mathematically proven.)

You'll need shielded ethernet cable that supports enormous bandwidths (that means, wavelengths from meters down to millimeters) to do terabit communications. That's physically not how twisted pair works; there's an inherent low-pass character to it that limits the usable bandwidth.

That's why we only have terabit/s communications over optical fiber: Whilst still electromagnetic waves, due to the extremely high carrier frequencies of light, dozens of Gigahertzes of bandwidth are achievable.

Note that doing high-rate electrical (e.g. on PCBs between your network card and your CPU, or even Gigabit ethernet) or optical links already needs basically all of the techniques that you see in wireless comms – synchronizers, equalizers, phase recovery, extensive channel coding, higher modulations¹.

Of course, the rates you get with a dedicated medium such as a fiber optic channel or a microwave guide on a PCB are higher than that in mobile comms, simply because of shorter distances (hence lower loss) and reliable channels (a lot of the rate of the channel in wireless comms is occupied to getting to know how the channel looks like, so that one can correct it at the receiver).

Your "factual" claim hence is plain wrong: You can certainly get higher rates over twisted pair then over some microwave link; 10 Gb/s ethernet over twisted pair is a thing. But the complexity of the receiver doesn't stand back behind that of a wireless receiver, in that case.

¹ And in fact, high-rate optical brings its own can of worms that's worse in at least one aspect than how we usually approach wireless comms: You'd drive the optical fiber in a power region that makes it a nonlinear medium; then suddenly, all of the precious assumptions about how different frequencies are orthogonal and how signals add up linearly break down, and that has catastrophic consequences for a lot of the math that we're used to.
Marcus Müller
  • 88,280
  • 5
  • 131
  • 237
21

So why is wireless transmission catching up to Ethernet in therms of transmission speed?

Wireless is not catching up to Ethernet in terms of transmission speed, but it is becoming fast enough that most people don't need to use faster wired systems. However, if you want to buy faster wireless devices, checking the prices ethernet modules on fs.com is instructive:

  1. 10gbit SFP+: $18.00
  2. 40gbit QSFP+: $39.00
  3. 100gbit QSFP28: $99.00

So to put that in perspective, a typical Intel branded Wifi card costs a few dollars less than an ethernet module with a real world bandwidth about 100 times greater. From that point of view, wireless is not very fast.

However, the real advantage of wireless in terms of cost is that the bandwidth is shared between all clients, whereas in wired ethernet, the bandwidth increases as the number of devices increases. This makes highspeed ethernet switches (which may require terabit/s switching speeds) extremely expensive. They also have hundreds or thousands of times more bandwidth.

user1850479
  • 14,842
  • 1
  • 21
  • 43
  • and cable / segment length ... in the same steps that speeds rise - the cable / segment lengths shorten (as those lengths directly depend on transmission speed) so in fact at some point we will have terabyte ethernet usable over 10 centimeters ... *laugh* – eagle275 Dec 16 '19 at 13:38
  • 2
    @eagle275 that's true for copper cables, but newer Ethernet standards are fiber and thus have very long range. The $18 dollar part I linked has 300m range, but a 10,000 m ($24) unit is still cheaper than a WiFi m.2 card. – user1850479 Dec 16 '19 at 15:45
  • 1
    FWIW 400Gbps is out there now, mostly experimental still but Cisco et. al. are picking it up. – Der Kommissar Dec 16 '19 at 19:11
  • 1
    The advantage of wireless you mention is also a disadvantage: a 40-port gigabit Ethernet switch might need to be provisioned with 40 Gbps of internal bandwidth, but that means that if everyone tries to talk at once, they all get the full speed. In comparison, a gigabit wireless access point with 40 users will provide no more than 25 Mbps per user if everyone tries to talk at once. – Mark Dec 16 '19 at 23:24
18

So why is wireless transmission catching up to Ethernet in therms of transmission speed?

Money. Time is money, but convenience is REAL money.

Because the speed-per-user cost is relatively low, and because the portability of wireless has a value almost beyond price. Yes, the infrastructure to get GHz data to hundreds (thousands?) of phones per cell tower simultaneously is really expensive, but the alternative - a fiber optic cable to every moving car on the highway - is flat out impossible. That guaranteed money flow drives what is, if you stop and really think it through, near-alien-technology development. Once the cell people have developed the tech, and the illusion of need, trickling it down to consumer home networks is relatively simple.

And again, running wires is hard. Especially in homes. Outside walls have insulation, inside walls have wiring, plumbing, and god know what else, and the cosmetics, esthetics, and budget are very different from what is considered normal for a commercial building. Think about the inconvenience of moving an ipad from room to room by unplugging and plugging cables. That's money, and that money drives astounding development. Using 1980's chips, a $30 home wireless router would be the size of a microwave oven and cost as much as a car.

AnalogKid
  • 16,865
  • 1
  • 13
  • 25
  • 2
    you know really big microwave ovens and pretty expensive cars ;) the linux-capable CPU in that router alone would be fantastic for 1989, not to mention the 64 MB of RAM or the 4 GB of flash storage, which would have been six 5.25" hard drives in 1989, weighing about as much as my microwave oven at home alone... – Marcus Müller Dec 13 '19 at 15:55
  • 1
    Yes that's about what I had in mind. The demand is so high that we do whatever is takes to make is work. Not sure if I should accept this answer, I'm afraid of confirmation bias :) – Francois Dec 13 '19 at 15:59
  • Well, the answer is excellent (hence my upvote), but it really neglects the physics that simply **cannot** allow terabit comms over twisted pair – see my answer. There's no "if I throw enough money and research at it" solution for Shannon Capacity – a link of limited physical bandwidth can only do limited transmission rate, that's math. – Marcus Müller Dec 13 '19 at 16:05
  • 1
    Embrace your biases. If you are over 12, they are who you are. – AnalogKid Dec 13 '19 at 16:19
  • @MarcusMüller How about terabit comms over a cable consisting of 1024 twisted pairs? – Hagen von Eitzen Dec 15 '19 at 14:01
  • @HagenvonEitzen actually, pretty hard! – Marcus Müller Dec 15 '19 at 14:11
  • 1
    @HagenvonEitzen You'll need excellent inter-pair isolation once you put that many twisted pairs into one cable (which is why Cat7 already has *per-pair shielding*; imagine the effort!), otherwise you'll get plenty of crosstalk at the 125 Mbd signalling rate of Gigabit ethernet. Then, you suddenly get a MIMO channel, quite like in multi-antenna wireless comms, but way more massively MIMO than what wireless comms currently do at scale. The same applies (but slower) to DSL – that's why VDSL2 has what is called *vectoring* in the DSL world (it's really cooperative MIMO). You **won't** get 1024× – Marcus Müller Dec 15 '19 at 14:15
  • the single-pair speed, but with a bit of luck, somewhere maybe around 2/3 of that (a wild guess based on very optimistic assumptions on the uncorrelatedness of individual signal paths). I'd presume less. – Marcus Müller Dec 15 '19 at 14:16
14

There's room for argument about some of the details. For example, I've omitted some of the rarely-used standards such as the original 802.11 and 802.11a, as well as some of the early 100 GBps Ethernet standards, in attempt at getting at least some approximation of speeds that were used fairly widely, not things that ended up as little more than technology demos. Depending on what you decided to include, you could get a somewhat different graph from this.

With that proviso, I get a graph of relative speeds something like this:

enter image description here

That looks to me like since the introduction of 802.11b, the two have been tracking fairly closely in terms of speed growth. Even though this is on a logarithmic scale, it looks like Ethernet (ignoring the pre-1995 period) has marginally faster growth, but given the small number of data points, I don't think either has a really massive advantage (note, that is to say: Ethernet has a speed advantage at all times, but its speed advantage has remained approximately constant for quite a while).

It's also worth noting that as speeds have gone up, distance has gone down (for both technologies). The newest/fastest 802.11 (802.11ad) operates at high enough frequency that its target range is "at least a meter", and generally won't go through walls at all. At least to me, that seems to leave some room for question whether we're likely to see it continue growing a whole lot. If I were being entirely fair about things, I'd probably have left it off the chart entirely--actual use seems to be extremely limited, at best. Even if I included it, I probably should have cut the speed in half (or so). What I've shown is the 8 Gbps variant, but most of the wireless routers I've seen only do the 4 Gbps variant (though most can also do an 802.11ac and 802.11n stream at the same time, so they're around 5.5-6 Gbps total).

One other note: this is all based on theoretical maximum speed supported by a given standard. In terms of real speed, it seems (at least to me) that 802.11 is typically slower by an even larger factor than the theoretical maximum implies. With Ethernet you can routinely expect to use 80% of theoretical bandwidth, but with 802.11 you're lucky to use 50% of theoretical (and often substantially less than that).

Jerry Coffin
  • 3,336
  • 17
  • 18
  • I'm not sure how you arrive at "I don't think either has a really massive advantage". Your few data points (which cover an eternity in IT) show that at any point in time wired was at least 100 times faster. – Peter - Reinstate Monica Dec 14 '19 at 08:22
  • 1
    I think he means in terms of one growing faster then the other i.e. the two curves are mostly parallel. – Francois Dec 14 '19 at 15:43
  • Yes, Ethernet is consistently fastest at any given time, but improvements in speed have been similar. – Jerry Coffin Dec 14 '19 at 16:10
  • Ah, the edit makes clearer what you wanted to say. Although I'd say that in the context of the word use by the question ("why is Ethernet losing its speed *advantage* over wireless") the answer should rather be "wired always had a huge advantage and is not losing it by any means". – Peter - Reinstate Monica Dec 15 '19 at 04:58
9

A major advantage of wired Ethernet for some purposes, to which wireless Ethernet can never catch up, is that the amount of bandwidth available in one cable will be unaffected by bandwidth utilization in other cables nearby. If one has 100 clients in a room each talking to its own server, and the cable topology allows every client to talk to its server without having to traverse any cables that are used by anyone else, each of those client/server pairs could have just as much bandwidth as if the other 99 didn't exist. By contrast, if the clients and servers were all in the same room trying to communicate wirelessly, they would only be able to use a small fraction of the bandwidth they would have available if they were alone.

A related advantage is that wired connections can deliver bandwidth reliably and smoothly, without hiccups. Although it's possible to stream video wirelessly with relatively short delay, it's not possible to guarantee 100% delivery of all frames within a short window. By contrast, if one is using a dedicated wired Ethernet connection between a client and a server, and if the client and server are themselves fast enough to handle the traffic, every packet will be delivered within a fraction of a millisecond from when it was sent.

supercat
  • 45,939
  • 2
  • 84
  • 143
5

Perhaps it's not about the networking at all, but the applications using the network. The applications currently running on these networks do not require more bandwidth. Consider, Netflix advises 25 Mbps for 4k video streaming. In the consumer space, that may be one of the hungriest applications. Since there are no consumer applications that need more bandwidth there is no evolutionary pressure to build networks with more bandwidth. In terms of straight download speed you are probably more limited by the pace of the content buffering on the other end than you are by your own network. Again, faster consumer bandwidth doesn't add any value. If someday we have applications that need more bandwidth, then I would expect both wired and wireless (which caught up to wired by then) to simultaneously increase their bandwidth to support those applications.

scorpdaddy
  • 1,177
  • 7
  • 11
3

There‘s one really big advantage of wireless over wired regarding attenuation.

A wired coax has ~X dB/m attenuation, whereas a wireless link has ~20 dB per tenfold distance attenuation (1/r^2).

This already pays off in favour of wireless for distances >100m.

This is why you could never do a wired cable link from earth to moon, but for a wireless radio link, NASA already did it 50 years ago.

Dave Tweed
  • 168,369
  • 17
  • 228
  • 393
Stefan Wyss
  • 6,471
  • 1
  • 11
  • 22
  • 2
    In counterpoint, the 60 GHz spectrum added in [802.11ad](https://en.wikipedia.org/wiki/IEEE_802.11ad) only has a range of meters, not tens of meters, and doesn't pass through most walls. – Eric Towers Dec 13 '19 at 23:49
  • 2
    "1/r^2" is really only valid in space. Terrestrial links have to deal with ground reflections, atmospheric attenuation and the curvature of the earth. – Peter Green Dec 14 '19 at 06:38
  • @PeterGreen Yeah, but terrestrial links often have antennas with high directivity. NASA used parabolic high gain antennas for their link to the moon. Also WLAN antennas provide a bit of directivity. – Stefan Wyss Dec 14 '19 at 10:55
  • 1
    The 1/r^2 reduction in power density is the best that be achieved. Once you're in the far-field, ~10 wavelengths away from the antenna, that equation holds no matter how tightly focused, or narrow the beam is. – SteveSh Dec 14 '19 at 17:05
  • @SteveSh You might be wrong about this: The 1/r^2 comes from the geometry of a sphere. What about a laser beam? Still 1/r^2? – Stefan Wyss Dec 14 '19 at 18:06
  • @Stefan Wyss - Agreed that it's related to the geometry of a sphere, But here it's about the surface area of a sphere. If an RF beam has a certain width in the x direction x (3 dB points on the gain contour) at d distance from the source, it's going to have a width twice that, 2*x at a distance 2*d away. The same applies in the y axis. So the surface area that the wavefront impinges upon goes by 4 for a 2x increase in distance. Hence the 1/r^2 relationship. Not sure about lasers. – SteveSh Dec 14 '19 at 18:26
  • @SteveSh Even a laser been is a cone after enough distance, isn't it? – Hagen von Eitzen Dec 15 '19 at 14:04
  • 1
    It only pays for it if you use electric cable where the attenuation is large, with optical fiber it is so much lower than it's not a problem (unless you're looking at distances where you wouldn't be able to use a cable anyway). – meneldal Dec 16 '19 at 07:32
  • @StefanWyss: For many lasers, the distance where you start to notice the 1/r^2 reduction is millions to billions of wavelengths. Still, that's meters. – MSalters Dec 16 '19 at 09:46
2

As others have noted, there are Ethernet standards for far higher than 1GB transmission. However, these have relatively small penetration into anything other than specialised datacentre installs.

The average home user, or business "office floor" user is still using Cat5 or maybe Cat6 cabling. The latter can theoretically do 10G, but in most cases probably only runs at 1GB. The reason for this is the cost of 10G is probably far greater than the real benefit of it. That is, the average desktop/laptop user, even the "power" users simply don't need that much bandwidth all of the time. For the few times in the month/year when they do, sure, it goes a bit slower, but the cost of that delay is less than the cost of provisioning 10G to them.

As noted elsewhere is that the higher "rated" bandwidths of wireless technologies are shared amongst all clients on that same network. Thus contention can dramatically reduce the real, available bandwidth to any given client. Further, there tends to be a larger protocol overhead for wireless networks compared to wired - so (say) a 1GB wireless connection, dedicated to one client will actually download data slightly slower than a 1GB wired connection (my rule of thumb is about 10% protocol overhead for wired, and 20% for wireless, but I'm doubtless waaay inaccurate there). Thus, higher bandwidths look good in marketing brochures, but don't necessarily mean that's what you'll actually get.

Just to labour the point a bit further... Bandwidth (and the equipment to provide it) is demand driven. Right now at least, the average user simply doesn't need the bandwidth of a simple Cat5/6 cable, let alone the additional bandwidth of later standards. Most users just need to download a few files, maybe watch some videos and send some emails. As such, "1G is enough for anyone", at least right now. I'm sure when we start getting full-brain implants that need more bandwidth, the cabling and connection standards will appear to support it, and with enough people to buy it, the price will be low enough that it'll become consumer technology.

All the while though, even the average household could completely consume many of the more common wireless standards - just by doing the "simple" activities I mentioned above with phones and laptops. Those people will absolutely want additional bandwidth, and so will be looking to upgrade to ever faster wireless networks for a few more years at least. Another solution for some of those people would be to use more Wireless Access Points to divide users amongst channels which don't affect each other. That more than doubles the install cost though, so the demand for higher bandwidth remains.

Ralph Bolton
  • 277
  • 1
  • 4