20

https://en.wikipedia.org/wiki/10BASE5 gives a succint but vivid description of the physical layer of the first version of Ethernet and its subsequent replacement:

10BASE5 (also known as thick Ethernet or thicknet) was the first commercially available variant of Ethernet. The technology was standardized in 1982[1] as IEEE 802.3. 10BASE5 uses a thick and stiff coaxial cable up to 500 meters (1,600 ft) in length. Up to 100 stations can be connected to the cable using vampire taps and share a single collision domain with 10 Mbit/s of bandwidth shared among them. The system is difficult to install and maintain.

10BASE5 was superseded by much cheaper and more convenient alternatives: first by 10BASE2 based on a thinner coaxial cable, and then, once Ethernet over twisted pair was developed, by 10BASE-T and its successors 100BASE-TX and 1000BASE-T. In 2003, the IEEE 802.3 working group deprecated 10BASE5 for new installations.[2]

Usually with information technology, the first version is simple and low-performance, and later versions are more complex, so the above is mildly unusual in that the first version was complex and high-performance.

But it is unequivocal that the coaxial cable of the initial version was superseded by twisted pair. Given that computers had been communicating over twisted pair since the sixties (with e.g. terminals, via RS-232), it seems surprising that there was a digression to coaxial before returning to twisted pair.

I remember in the old days, it seemed like the bus topology was the simplest and most efficient. I was surprised when later, the star topology turned out to be more so.

Did it also seem that way to the engineers who designed Ethernet? Is it a case of 'yes, we had to try it that way to find out it wasn't the best option, but if we were doing it over, we would go with twisted-pair star topology from the start'?

Or is it a matter of bandwidth? The electronics of the time could not deliver ten megabits per second over twisted pair, and a separate coaxial line to every client machine would have been too expensive, leading to the bus topology?

Or is there some other consideration I haven't thought of?

15
  • 3
    Ethernet (don't know about 10BASE5 specifically, but Ethernet itself) was developed for use as an embedded LAN inside of Xerox's photocopiers. Coax was probably just fine for that application - not only that, but better than fine given the (probable) electrical noise inside of those boxes. Also, to be able to connect those copiers to workstations in their building. A big corporate campus building. Running coax was not a problem - not compared to the capabilities it provided. (Copiers/laser printers/and the like were damned expensive back then!)
    – davidbak
    Nov 13 at 0:03
  • 5
    @davidbak Do you have a source for that? Wikpedia's Ethernet History section gives the history as I know it, which is that it was invented in PARC, a computing research lab that had little to do with photocopiers. And "an embedded LAN" in photocopiers doesn't make much sense; why go to all that extra effort when you could use much simpler serial and parallel connections? (BTW, coax makes perfect sense when you want a wire "ether" to replace the radio ether, since when terminated it functions electrically in the same way.)
    – cjs
    Nov 13 at 0:43
  • 4
    @cjs - source: my brother, who worked for xerox parc. and if you think parc had little to do with photocopiers / laser printers / and the like, think again. also it does make sense - wiring harnesses cost money - especially in support/maintenance. Though now I think of it ... that project he was on to replace wiring harnesses in photocopiers / laser printers with ethernet (which did happen) probably was much after the invention of ethernet.
    – davidbak
    Nov 13 at 0:47
  • 3
    @davidbak I know PARC made laser printers from photocopiers in the '70s. That they'd get involved with making photocopiers per se in the '80s makes some sense, since that was about the time things reversed and they started making photocopiers from laser printers. The same goes for intra-device LANs; they made little or no sense in the '70s, but started to in the '80s as the power of individual components increased and LAN interface chip costs decreased. Thus, CAN bus and the like. But this was not what Ethernet was originally developed for.
    – cjs
    Nov 13 at 1:49
  • 7
    Back when we started using 10BASE5 in the Computer Science department where I went to college, we didn't think it was a mistake. We thought it was ****ing amazing! I had a friend at another school who flat-out refused to believe me when I tried to explain it to him. His professors had him convinced that communication over more than a few inches at those speeds was physically impossible. Nov 13 at 17:14
42

The performance (except for allowable cable length, see below) was the same between 10BASE5, 10BASE2 and 10BASE-T, and you have the complexity backwards: the coax is simpler than the twisted pair.

The key point to remember is that Ethernet uses a shared transmission medium (much as radio does) that can carry only one message at any time. Thus, a transmission must be received by every station and reception must start very soon after transmission started. (It won't be instant because the speed of propagation in the media is not infinite.) Further, for collision detection, stations must be able to listen at the same time as they transmit so they can see when their message is being garbled by interference from another transmission.

For both 10BASE5 and 10BASE2 the transmission medium is extremely simple: you have a single transmission line, terminated at each end, and stations are connected to the transmission line at arbitrary intervals. Note that the transmission medium itself is entirely passive, consisting of just a cable and termination resistors at each end. (The termination is required to prevent reflections that would introduce what look like additional signals to receivers.)

10BASE5 had a maximum length of 500 meters. 10BASE2 reduced the maximum length to 185 meters, but where that much lower length was sufficient this was a good trade-off for having cable that was cheaper and easier to handle. In part this was due to 10BASE2 cable being thinner, but the really big advantage was that, while the 10BASE5 medium was a single long cable, the 10BASE2 medium could be constructed from many segments of short cable, joined together with "T" connectors at each station. This allowed much easier re-routing of the transmission medium to include, e.g., a new room off to the side of the existing run.

The 10BASE-T transmission medium is more complex and, at the time it was introduced, more expensive. Rather than a simple passive cable, it consists (originally) of one or more hubs to which the stations make point-to-point connections over twisted pair. These hubs are active devices (to at least some extent—a lot more complex than a cable and couple of resistors anyway) that are responsible for "synthesizing" a shared medium with correct electrical characteristics via their connections to the stations and each other. (To be clear, you could not simply take a bunch of 10BASE2 cables from stations and connect them all together in a star shape at a central point; the electrical characteristics of the transmission medium would be wrong and it would not work reliably, if at all.)

Twisted-pair systems grew considerably more complex yet over time. From the early 2000s onward hubs had pretty much been entirely replaced by switches, which are effectively a second station at the other end of the link from a user station: they send and receive full packets regardless of what's currently on the shared medium and, when they receive a packet while the shared medium is busy they store it for later forwarding. Features such as full duplex station connections (so that a station could send and receive at the same time) and optional higher speeds introduced the need for negotiation protocols to allow the station and switch to agree on what was bilaterally supported.

These days, with electronics having fallen by many orders of magnitude in cost, the cable cost has become a much larger component of overall system cost. (Twisted pair cable is much cheaper than coaxial cable.) Thus, twisted-pair systems are cheaper than even 10BASE2 in most practical applications. But it may well be the case that for an installation of, say, three computers close together, a coax cable, three T-connectors and a pair of terminators is cheaper than a four-port switch, which still runs around $10 at retail. (That does raise the issue of how to price the 10BASE2 cards, however, which have not been manufactured in a long time.)

12
  • 5
    @jpa I was so intrigued by this that I popped over to Digikey to have a look. But looking at 1000' quantities, CAT5E seems about half the price of RG58 (And of course CAT5E is good for about ten times the data rate.)
    – cjs
    Nov 13 at 11:03
  • 3
    @cjs - Where are you getting that data rate comment? CAT5e is 4 pairs that are specced up to 350 MHz. Some varieties of RG58 can go up to 5 GHz. Of course, there are additional issues (how you're doing the signal encoding), but at least the pure channel capacity of coax is much higher then cat5e.
    – Fake Name
    Nov 13 at 12:51
  • 2
    "stations are connected to the transmission line at arbitrary intervals." - it is my understanding that the connections had to be at specific locations on the coax. Signal interference perhaps. Nov 13 at 13:55
  • 2
    @ThorbjørnRavnAndersen For 10BASE5, yes — it was very sensitive to reflections (same reason why it couldn't withstand T connectors). 10BASE2 just had a minimum distance between tees (which was small enough to not be much bother).
    – hobbs
    Nov 14 at 6:24
  • 10
    "Performance was the same ... 10BASE-T": this is missing the big qualifier of "eventually after a decade of more development". Many people seem to be under the impression that these all came out about the same time, but it was a long wait to get to ethernet on twisted pair. There is no question that twisted pair had huge challenges in maintaining signal integrity and dealing with interference. That's why StarLAN-1 (first ancestor of 10BASE-T) was only 1Mbps
    – Grunthos
    Nov 14 at 17:34
19

Something that enabled functional multiaccess local area networks hardly seems like a mistake.

At the time, or at least in my experience, your networked computers were in a dedicated room. The cables and vampire taps were awkward, true, but manageable in context.

I actually had a fairly cynical view of 'cheapernet' (thinwire Ethernet), at least if it was daisy-chained through a number of cubicles. Someone upstream could disconnect the wrong wire and bring down the chain. Maybe I just worked with the wrong sort of programmers; this was also about the time when they were moving from 'terminal on a desk' to 'small VAX on a desk'.

Twisted-pair was superior in that no-one tried to daisy-chain it.

There was about a decade of useful lifetime between original Ethernet and 10BASE-T, so I'd say "not a mistake".

4
  • 2
    "Twisted-pair was superior in that no-one tried to daisy-chain it." - IBM did with their token ring products. They overcame the issue where one station could bring down the entire network with a complicated connector and really switch. Nov 13 at 18:33
  • 1
    I meant "token-pair Ethernet", of course. But yes, token ring was a thing, back when IBM claimed that CSMA/CD was no good because it could not guarantee fair access to all stations. Nov 13 at 20:55
  • 1
    There's a new standard coming out for industrial use, 10BASE-T1L, which does a token-ring-like ETH over a single twisted pair, with something like 1.7km max cable length. Another thing popular in industrial devices is faux daisy chain - 100BASE-TX device with built-in switches. Nov 15 at 11:27
  • 1
    Yeah, a number of manufacturers make 3 port switches that are designed to essentially replace the PHY chip in a device. So the extra cost of adding a daisy chain port nowadays is fairly small. Commonly seen in IP phones too. Nov 16 at 6:33
10

Coax had the advantages of high bandwidth (twisted pair wiring at that time was only good for 1Mbps) and a simple collision distributed collision detection scheme for access control.

The collision detection mechanism for 10Base5 involved detecting the DC level on the coax. When transmitting the attachment unit injected the manchester encoded signal with a bias of 40mA that results in 1V DC on the cable (50 ohm termination at each end) if a second unit transmits the voltage goes up to 2V. Each active unit senses that voltage and if over 1.5V determines there is a collision and will then backoff.

About 1985 AT&T was promoting a 1MBps system based on ethernet but using standard twisted pair phone wiring. That required a hub to propagate the data and implement the collision detection. Starlan never caught on.

LattisNet was the first 10Mbps over twisted pair network that supported 10Mbps about 1987. It evolved into 10BaseT over the next few years.

Ethernet switches became available in the late 80's, I think Kalpana was the first company to introduce a 6-port one.

1
  • 4
    StarLAN as a brand didn't catch on, but the technology absolutely did. StarLAN-1 morphed into StarLAN-10 at 10Mbps. StarLAN-10, plus the addition of link beat, is 10BASE-T. LattisNet was a competing implementation of 10MBps EoTP that did indeed have good initial commercial success, but it was not the basis of 10BASE-T. (Some network interface cards such as the 3Com 3C-523 could be used with either StarLAN 10 or 10BASE-T, by switching link beat on or off)
    – Grunthos
    Nov 14 at 17:44
4

The other answers leave out a hub-like mechanism to attach 10BASE5. Often the thick yellow cable was an expensive asset, connecting one more vampire clamp wrongly could ruin it.

The computers and the vampire clamp were connected by AUI cables, with 15 pin D-shell connectors, having a locking slider. So hubs were invented with multiple AUI ports, e.g. 8 incoming and 1 outgoing, where just anyone could attach one more computer. Whenever the outgoing port was left free, this worked like the twisted pair systems later: plugged connections, a single collision domain on the net.

For installations where the long distance of yellow cable was not needed, link inside a computer room, these multi-AUI-connectors could even be cascaded.

For combining AUI and other systems, adapters were introduced. They were made simple by already having a power source through the AUI cable.

In conclusion, all the existing variants came one after another, when ever engineers had more ideas how to scratch an itch with the previous ones.

1
  • 1
    One problem with the multiport transceiver is that they changes the rules for 10Base5 with regards to how many hops a signal can take. Normally a Ethernet network can have 4 repeaters in any path (ie 5 cables connected via a repeater) but a multiport transceiver costs one such hop (or about there.) This due to demand that the maximum propagation time isnt' allowed to be above a certain number of milliseconds (this due to how collision detection works.) Nov 19 at 9:11
1

Usually with information technology, the first version is simple and low-performance, and later versions are more complex, so the above is mildly unusual in that the first version was complex and high-performance.

I think I would dispute that characterisation. The early versions of Ethernet used a coaxial cable with drop cables. That's pretty simple and coax had been in used for years. For example, the connection between our TV and the arial on the roof of our house used a coax cable.

Later versions of Ethernet shrank the length of the coax cable until it fitted into a box (called an ethernet hub) and increased the length of the drop cables. All the complexity is still there, it's just hidden inside one box instead of distributed over your entire office.

But it is unequivocal that the coaxial cable of the initial version was superseded by twisted pair. Given that computers had been communicating over twisted pair since the sixties (with e.g. terminals, via RS-232), it seems surprising that there was a digression to coaxial before returning to twisted pair.

Not really. Consider that Ethernet had to operate at speeds that were several orders of magnitude higher than the typical dumb terminal of the 70's. The cables that terminals used to communicate via RS232 were really not close to being up to the job.

And the coaxial cable was not superseded by twisted pair, it was superseded by hubs.

I remember in the old days, it seemed like the bus topology was the simplest and most efficient. I was surprised when later, the star topology turned out to be more so.

That depends on what you mean by simple and efficient. Firstly, it should be pointed out that Ethernet is still a bus topology despite appearances. It is simple in some respects. A computer on Ethernet simply waits until nobody else is talking, broadcasts its packet and listens back to make sure there is no collision. Star topology implies a computer in the middle to mediate all the communications. Ring topologies require a token to be passed around the ring.

Traditional coax based Ethernet is not simple in physical terms though. It requires a coaxial cable to run past all the computers on the LAN segment. This was acceptable whilst computers were in labs, but once you start putting computers on every desk in your entire office, it becomes untenable.

1
  • 1
    The cables that terminals used to communicate via RS232 were really not close to being up to the job. Though the opposite was quite common. I wired up plenty of serial terminal connections in the 90s on CAT3. Nov 17 at 13:21

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.