12

In a discussion on the history of Ethernet and 10BASE5, Stefan Skoglund commented Was 10BASE5 a mistake?

One reason why 10BaseT became possible is Moores law (and the same for the other designs after that.) The designer's chip budget was only good enough for this.

This is an interesting remark that I think is worth expanding on.

Does it take more sophisticated electronics to drive Ethernet over twisted pair than coaxial? If so, is there a way to quantify that? What's the reason for the greater difficulty? What sort of transistor count is needed?

Ethernet ran at ten megabits per second, which was an extremely aggressive speed for the seventies, and even for the eighties. (I remember around 1990, seeing a 286 PC read data faster from a server, over 10Mb Ethernet, than from its own local hard disk – in other words, 1970s Ethernet speed was arguably still overkill in 1990.) Given that RS-232 was running fine over twisted pair since the sixties, presumably whatever difficulty there was, was not so much 'difficulty of communicating over twisted pair' as 'difficulty of communicating over twisted pair and still maintaining signal integrity at ten megabits per second'?

6
  • 2
    For what it's work, just because coax 10Base5 internet ran at 10 Mbit/sec doesn't mean that computers could deal with that much data. I remember when we got our first Ethernet card for a VAX 750. It was huge (bigger than an S-100 card, if I remember correctly). I think it was somewhat less than CDN$10k. It ran at (if I remember correctly) at 1/4 the speed of the network (i.e., it would run at 10 Mbit/sec on the cable side, but a quarter of that on the bus side.
    – Flydog57
    Nov 20 '21 at 4:10
  • 8
    Also, given that, before it became cost-effective to throw switches everywhere, the network was a shared medium, you want it to run faster than the machines if you can afford to, so you've got capacity left for multiple interactions between different machines to share the medium... like time-division multiplexing but less regimented... same reason you want to check the backplane bandwidth on a switch. Those cheap gigabit ones tend to have 10Gbit backplanes whether you buy the 5-port ones or bigger.
    – ssokolow
    Nov 20 '21 at 4:38
  • 3
    ...meaning that, aside from different strategy for dealing with the potential for collisions behaviour, a cheap gigabit switch is analogous to putting all computers on the same switch on a shared a 10GBase-5 trunk if such a thing existed.
    – ssokolow
    Nov 20 '21 at 4:44
  • 1
    I do not think that it takes more sophisticated electronics to drive Ethernet over twisted pair (literally taken), but the main difference is that TP is always point-to-point, and so you need more sophisticated electronics to make a TP only network behave in the same way as a coaxial cable network.
    – chthon
    Nov 20 '21 at 8:05
  • 2
    Also, the electronics design in the 1980s and earlier tended to be kind of conservative, the idea of using twisted pair cable (which everyone knew mostly as a baseband phone cable) vs coaxial (which was proven already) for a high frequency application would probably have appeared outlandish.... Nov 20 '21 at 15:43
25

I designed a 10BASET ethernet chip (codename ENZO used in Sage MainLAN https://archive.org/details/byte-magazine-1991-01-rescan/page/n67/mode/2up?q=sage+mainlan) and produced a proposed design for a 10BASET hub.

In electronics terms 10BASET is simpler, a pair of resistors for pulse shaping the output externally, a transformer for isolation and for the input side I have now forgottem I'd guess some sort of op-amp.

There is a little more stuff required inside the ethernet chip set to generate link test pulses and the pulse shaping signal, but we did save having to have DC<->DC converters for the MAU and the 8392 for 10BASE2.

20

Your question seems to ignore the topology of a network completely:

If you want to connect, say 20 computers via Thickwire or BNC, you need:

  • A length of the chosen cabling
  • 20 MAUs if you want Thickwire
  • Some cheap termination resistors

That is: No active electronics except the MAUs (which you could consider part of the computer's Ethernet card)

If you wanted to do the same with 10base-T you needed:

  • Roughly 25 pieces of 10-base-T wiring
  • At least 5 four-port hubs (or one single 20-port hub which was a really expensive beast)

The point is: You didn't need more sophisticated electronics for 10BT, but you needed more of it. 10BT wants active devices at each branching point, coax doesn't.

5
  • 4
    (5 five-port hubs for 20 computers because you need ports to connect them together too.)
    – ilkkachu
    Nov 20 '21 at 17:22
  • 1
    @ilkkachu I was pretty sure someone would come along and nitpick ;) That's why I wrote "at least" .
    – tofro
    Nov 20 '21 at 23:41
  • 1
    hehe, sorry, I couldn't help it. But it just works to prove your point. Lower port count would be even worse, I tried to count how many 4-port hubs would be needed for 20 computers, and couldn't do it with less than nine.
    – ilkkachu
    Nov 21 '21 at 12:20
  • 2
    If you connect the 5 4-port hubs with Thickwire (wouldn't count that as a "port"), you might prove the point again.
    – tofro
    Nov 21 '21 at 17:31
  • 1
    If one counts the number of transceivers circuits, a 10-base-T network requires two transceivers for each connected endpoint, plus another two for every hub beyond the first, while a network using coax would simply require one transceiver circuit per endpoint.
    – supercat
    Nov 22 '21 at 19:00
17

The per-port electronics in the network card is very similar, no matter whether you use 10base2 or 10baseT. AUI is slightly simpler on the network card, as a part of the electronics is not in the card, but in the MAU.

The primary difference between coax and TP is that TP needs a hub or a switch, which is an additional electronic device and includes a second transceiver per participating machine in the hub (or switch), whereas the coax cable is a completely passive device. The success of 10baseT started coming in when hubs got cheap enough that the advantage of being able to drop single misbehaving stations or broken cables outweighted the installation costs.

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.