Ethernet at 40: Its daddy reveals its turbulent youth When Bob Metcalfe, the prime mover behind the invention of Ethernet, recently visited the site of that invention, Xerox's Palo Alto Research Center (PARC), The Reg had the opportunity to sit down with him to discuss the history of Ethernet, its advantages over Token Ring, and IBM's perfidy. Metcalfe was in town to promote a …
I have known RM since the early 1980's - I sold 3com their first 100 PC's... Later, when I was a principal engineer at a major Boston area software company, we used to meet for dinner before IEEE meetings in the area (usually at Mitre). Now, I am a senior systems engineer at a tier-one mobile phone manufacturer - Robert and I are about the same age. So, I can only say ... "Robert! It is time you get back to your roots!".
P.S. I have utmost respect for RM. He has been one of my tech heroes for 30 years now. :-)
Yup - I remember soldering terminators because the "official" ones were ridiculously overpriced. Provided people kept their fingers off the cabling the stuff was actually quite stable, and the cards were pretty reasonably priced too (except the 3COM one we stuck in the server). However, try to set up a WAN and it got very expensive..
God knowns how many miles of coax I've used in those days but it was a *lot* :).
Once upon a time 3 Com shipped a customer a whole bunch of 10 meg ISA BUS Ethernet network cards, about 200 if I remember correctly, and these were fitted into shiny new 486 processor boxen and duly connected to 10 meg Ethernet hubs and switches for their data processing folks who were moving into a nice new office.
When it came time to test, nothing seemed to work correctly and even the tier 3 engineers were stumped by the fact that pings and trace routes would go nowhere. Much finger pointing and recriminations ensued until someone brought in a Network General sniffer and did some packet captures.
It turned out that all of the 3 Com network cards had been accidentally burned with the same mac address and that mac address was the only one that showed up in the packet traces.
I was a remote technician on maintenance support for the hubs and switches and was just as amused as everyone on our team that the problem was not ours, good times.
As a freshly graduated pup in the late 70's, I worked at a British university computer dept doing research into token ring fault tolerance. Our particular ivory tower had strong links with Cambridge University Computer Laboratory and its eponymous Ring (I once had the privilege of meeting Andy Hopper and Maurice Wilkes - yay!). Of course, that meant I was indoctrinated with all of the "ethernet bad, ring good" propaganda. After finally escaping to a proper job, I had to set up a network that functioned for real and ended up with ethernet. I still can't believe all that wasted time fscking about with s0dding rings.
Don't get me wrong, the ring was very clever and put together by some extraordinarily smart people, but ethernet was better in most respects.
(Ok, maybe it wasn't all wasted time, if one considers the valuable education)
Cue downvotes for nostalgic lovers of The Ring...
"......Don't get me wrong, the ring was very clever and put together by some extraordinarily smart people, but ethernet was better in most respects....." Agreed. I was introduced to Ring, loved it, it just worked, but it was very sad to see the lengths to which the ant-Ethernet crowd went to. Ethernet simply was better, but the technical discussion got hijacked by business people that saw Ethernet as a threat to their incomes, and then they put monetary pressure on technical and academic people to join one camp or the other. A lesson for those that like to cast M$ as the Big Bad in the PC wars - they were just following the business practices set by IBM.
There's no central authority instructing the nodes to act; they discover whether it's safe to broadcast through their own local observation. The lack of a centralised actor and the ostensible resulting chaos leads to a more efficient overall system.
I'm not a libertarian but I can see there's a reasonable argument in there.
Ayn Rand was a neo-fascist
Downvoted from the incursion from the left-winger's pigpen.
Ayn Rand was not a good writer and crazy radical, but had all the normative ethics written down for analysis. She detested socialist leechers and failures, in particular the socialist failures resorting to violence, so how can she be "neo-fascist"? What does that even mean?
Hold on, are you trolling?
It's not so easy to draw a political moral from the development of the Ethernet. If it's libertarian, it's more social libertarian ala Noam Chomsky rather than the capitalist libertopia of Ayn Rand. The nodes are just communicating with each other, after all. They're sending out packets, not accummulating them.
The only connection to Atlas Strugged is this. IBM tried to go Galt. They failed.
Rand wants everyone to be selfish and only follow their own goals because she doesn't understand that there is a difference between individual goals and the common good. Ethernet is actually a good example why she is wrong.
Ethernet works, because there are strict rules everyone must adhere to. If there is a collision you have to wait a random time. With each collision the timespan of that random number will increase. This rule throttles down stations when the network is full, to share the bandwidth more or less fairly between them.
Now an Any Rand network card would simply ignore the standard and re-transmission the packet right away. That way it will always give the card 100% of the network if it wishes to. However if more people act like this, there will _only_ be collisions. Once there is a collision, both stations will try again at the same time, they will do so until one of them gives up. Instead of sharing the bandwidth fairly between the stations, those few stations ruin the network for everyone.
In real life Ethernet worked very well, most networks had a few file servers that sent out most of the network packets, clients did not request the next bit of a file until they have got a response to the last request.
So yes, Ethernet has no central authority, but in real life it was very common for 1% of the machines on a network to be sending out 80% of the packets. DEC published a research report on “real life performance of Ethernet” that showed how much better Ethernet worked then you would expect if you modelled it with random unrelated packets.
What we call “Ethernet today” is nothing like what Ethernet was like, as today it is all switched, in the old days all the computers connected to the same cable using T-Connectors; there was no electronics between the computers, just plain cable.
Yes, I noted how this aspect of the evolution of 'Ethernet' was totally glossed over.
The reason why 10Base-T LAN is called Ethernet is largely down to marketing and lobbying of the IEEE by interested parties.
There were debates within the IEEE (and contributing organisations) as to whether 10Base-T really was an extension to 802.3 or should be a new 802.n standard. What seemed to finally swing the case for 10Base-T to be an extension of 802.3 and hence be called 'Ethernet' was that 'Ethernet' had a good market presence and image and hence vendors would find it easier to gain market acceptance of 10Base-T solutions if they were branded "Ethernet", along with the weight of agreement among the contributing members. Once the decision was made, the rest was history...
I suggest that 802.11/Wi-Fi is the only other 802 standard to have gain a similar level of market awareness.
"Multipoint data communication system with collision detection U.S. Patent #4,063,220
Everyone does this. You really think the difference between a low end smartphone and a high end smartphone is hundreds of dollars?
I remember having an Amstrad Fidelity satellite receiver which didn't work. Someone sold me a similar model that did work but didn't have the remote control. I opened both of them and found that all I needed to do was swap the control boards over. The cost difference was nothing, but there was probably a £50 difference in price.
Actually I think you'll find one of the differences between £70 and £400 cameras is also in the quality of the lens and zoom mechanism, the quality (not the pixel count) of the sensor to allow lower light shooting, the intelligence to drive the auto-focus at a decent speed, etc etc etc. A camera, like any techy device contains many separate parts that can influence the final product. No point having the worlds best sensor and autofocus systems and then throwing in a cheap plastic lens system.
Same chips and sensors maybe, but did you stop to consider that the lenses could be rather different?
A gazillion megapickels is wasted if the optics are not good enough. That's why a 6Mp photo taken with a Minolta Dynax 5D can be printed as an 8"x12" and look nice, but an 8Mp smart(arse) phone photo is barely passable as a 6"x4" print.
@Dave Ingram; in that case, it's also because the sensors *will* be different- same amount of pixels, but the one in the smartphone is likely to be smaller (and cheaper) with less light-gathering capacity per pixel, and hence more noise and more noise-reduction required compared to the one in a typical DSLR.
Some mainframe products from various manufacturers had optional extras that merely needed a single link on the backplane to be wire-wrapped to activate them. Intermediate mainframe models had added hardware delays to slow them down. It could even be a case of just loading a different firmware to effect a performance upgrade. They were all ways to standardise the manufacture - and the customer paid for a particular performance.
I setup a little business at Uni buying single sided FDDs and selling them on as double sided. All you had to do was open it up and reconnect a disconnected cable. My profit was half the difference between the priced the company sold the two versions.
Icon for the beer the little sideline kept buying for me.
Wrote :- "IBM produced a printer .. standard .. and advanced (more expensive) ... the only difference being moving a belt on a drive wheel to speed up the printing."
Anoher example was Windows NT Desktop vs the far more expensive NT server. The CD contained exactly the same code, except for a flag somewhere in the installation script that allowed the server features to be installed or not.
It is also fairly common to have a lower quality product when the production lines are getting up to speed, especially for things like graphics cards.
The ones that pass QA get sold as the top end product. The ones that fail QA - but still work - get sold as the varying grades of cheaper product.
The ones that fail QA and don't work get sold for OEM built in cards :p
If (as he says should maybe have happened) Simonyi had written Ethernet instead, we'd have a well-functioning open standards wysiwyg text editor, and an expensive (even more than TR) proprietary network interface that has Clippy the Paperclip trying to talk us into doing stuff?
/The good half of that outcome is more or less there (hello, LibreOffice4.0).
I know almost everyone says how wonderful twisted pair cabling was, but one of the major advantages of ethernet for us over Token ring was that it didn't require star wiring for every damn device. By and large we weren't installing in purpose built office buildings but in older, often historic buildings with no cable space. Getting permission for a run of thinnet, maybe with a thicknet backbone, or even earlier corvus omninet which was a STP bus, was straightforward, getting permission for vast wodges of twisted pair for TR would have been next to impossible. And then anytime anyone wanted a new device it would have been get the cablers in. Mind you that hasn't changed, because every time we think we've flood wired an office they have a reorg and cram yet more tinier desks in at an even higher density...
A few years on of course the managers all had their own devices on their desk and those pig ugly runs of trunking suddenly weren't a problem any more... Especially if it meant no more engineers crawling round our office with a TDR trying to work out where the break actually was...
To be quite honest the TDR wasn't nearly as useful as you'd think it ought to be. Because you never knew how many lengths of thin net had been chained together and wrapped round various desks a distance to break wasn't that useful. It was often just as quick to just run through the offices with a spare terminator.
Best thing to have was when Jane and I were throughly debugging out a rather chaotic install in the Fire Service HQ - a Victorian mansion on about 7 diffeent levels - and we managed to borrow a couple of walkie/talkie radios. "Ok, which room are you in now" "Terminator off NOW"...
I bet I've still got an animal made of thin net connectors somewhere [grin]
Ugh star wiring - surprised no-one mentioned it - for me the main killer of Token Ring.... I used to work in the insurance company Commercial Union who lived, ate and breathed IBM. They backed Token ring over Ethernet, OS2 over Windows, Smartsuite over Office, Lotus Notes over... well any other half decent email programme. I set up my own little enclave with a peer to peer Windows for Worksgroups network (remember that!) running over Ethernet, using Office.... I fought off battles from my IT Department...
Then about 2 years later the nerds in the IT centre panicked and came to us as the most experienced people using these technologies.... ah those were the days :-)
Wrote :- "I set up my own little enclave with a peer to peer Windows for Worksgroups network (remember that!) running over Ethernet, using Office.... "
Windows for Workgroups?! Nostalgia for that that PoS? Please tell me you're kidding.
In fact I had WfW on an old PC until just 6 months ago - I used it with an old but very fast scanner that had no drivers for anything later.
At least by the time Ethernet supported twisted pairs, you could support additional devices on a single pair just by using a very CHEAP “repeater”. So it was more a star of stars of stars setup for Ethernet.
You could get devices to do the same for Token Rings however they were VERY expensive
The joy of accidental linkages to dirty earth from exposed connector metal; threading a long AUI cable through a false floor - then discovering it was the wrong way round. Bee-stings that vibrated when a jet took off nearby. Badly engaged AUI slide-locks causing connectors to hang off - so they transmitted but didn't receive.
The thin coax was prone to user interference: kinked cables; disconnected cables; extra terminators; different impedance cable sections; departments installing "private" bridges that put a loop into the Company Wan. User software - usually a cloned install or config file - that used the same private MAC address to override the hardware one.
Some units had a deviant clocking rate that other suppliers' units couldn't sync with. Excessive dribble bits also caused compatibility problems. Some suppliers modified the back-off algorithm to give their kit unwarranted priority on a collision - so their performance measurements under high loads were better than the competition. Too many people thought that any "collisions" were indicative of a network fault - it should have been called "contention".
Initially the 10mbps technology at the physical layer was cutting edge, and expensive, before standardised Manchester encoding components made it easier. That possibly slowed down the initial market penetration.
The cabling advantages of 10base-2 went away with 10base-T, which went back to a star topology. It was educational to watch the utilization lights versus the collision lights on the hubs, behaving exactly as the theory predicted. Also, most original T-R cabling was done with Type 1, which was STP. It worked very well with Ethernet, either with baluns to terminate 10base-2 or adapters to terminate RJ-45. Installations going from T-R to 10base-T didn't have to replace their building wiring.
Back in 1991 I built a pair of lines for duplicating disks. Each line consisted of a server PC and 10 clients which pulled files from the server. One line was linked by 4Mbps T-R, one by 10Mbps 10base-2. For 1-to-1 duplicating, the T-R line took 12 minutes and the Ethernet line took 10 minutes. For 1-to-10 duplicating, the T-R line took 20 minutes and the Ethernet line took an hour and 20.
Eventually, switches replaced hubs and T-R's main performance advantage evaporated.
I'm not sure about Ethernet being better in the early days. Cheaper and simpler, perhaps, and beaconing was a bad thing, but troubleshooting non-star wiring breaks was horrible. Before switches, collisions were a problem which meant the headline Ethernet speed meant little on a congested network.
The joy of Ethernet was the ad hoc doom party - disconnecting from the office lan was easy, or at home nothing more than a bit of thinnet was needed.
The class idiot at college who would steal a terminator or disconnect one of the computers thus taking down the entire network. Everyone would then have to hunt to see where the break in the network was.
+ idiot system admin that insisted Windows 3 was loaded from a network drive. So each computer would take about 5 minutes to load windows on a bad day.
Glad to see the back of that system!
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
