Turing Award goes to Robert Metcalfe, co-inventor of the Ethernet Professor, engineer and namesake of Metcalfe's law Robert Metcalfe is the latest winner of the Turing Award for an invention he made back in the 1970s: the Ethernet. Xerox's Palo Alto Research Center (PARC) was just a few years old when Metcalfe arrived and set up the lab with that internet precursor, ARPANET. But that didn't …
At uni in 96/97, our lecturer was adamant that Ethernet was going to die off and be replaced by ATM networking. What I think changed was that Ethernet moved from the thin-net coax with all its myriad of problems with packet collisions, lost terminators etc and moved to a fully switched and duplex configuration, making it reliable and far less prone to collisions.
I also find it amusing that it was based on Aloha net which was wireless into a wired protocol and now it's come full circle into a wireless protocol again with Wifi.
Well done to Robert Metcalfe and all those who have had a part in developing and maintaining something we all rely on day to day!
Nothing changed. Your lecturer had drunk the ATM Kool-Aid. I worked in the industry at the time. Ethernet was already doing 100MBs over Cat5 cable when ATM was being talked about. It wasn't that Ethernet got better, it was that ATM was never, ever, going to work in a LAN scenario when it could only manage 25MBs at about 10 times the cost. When you added the behemoth Kludge that was LAN emulation, it barely struggled along at 4MBs.
The problem with ATM was it was supposed to be a technology capable of running WANs and LANs with one protocol. As such it was inevitably bound by the limitations imposed it by the need to be a WAN technology. (54 byte packets, anybody?). Ethernet did not suffer that limitation and that's why it won on the LAN. Gigabit ethernet appeared not long after that and every company that was trying to make ATM LANs a reality promptly gave up.
>(54 byte packets, anybody?)
53 bytes (well 48byte packet and 5 byte header) cos even numbers are for squares.
It's perfectly sensible, the datacoms guys wanted larger 64byte for high speed efficiency and the telecoms guys wanted smaller 32byte for better latency - so they decided on a perfectly reasonable compromise of 48bytes. then they added a separate header so you wouldn't have to bother reading the packet to find out what to do with it. then the committee went for a series of long lunches
I recall reading a book on networks by Tennenbaum*, I think, where he claimed that Ethernet was designed by people who wanted something that actually worked, rather than 'just another standard' that no one would use. I think it is safe to say Metcalfe and Boggs were successful.
Cheers!
*Well, bits of it, not the whole thing, although it was well written, it was a bit too technical for me at the time.
I'm currently using a 150 ft length of orange (plenum) ethernet cable as an antenna feedline. Works great at HF.
(and N connector >> vampire tap) At Data General, we very quickly learned that crimp on N connectors and 3Com transceivers were the reliable way to go. And I still despise the slide latch AUai connectors.
Way back in the day at CMU, I actually used the original 3 Mbit Ethernet, due to CMU's having received a grant from Xerox PARC of a dozen or so Altos, a Dover laser printer and a bunch of Ethernet cards. My recollection was that the Ethernet was just a long Cable TV cable, and each machine on it would tap into by drilling a hole through the outside of the cable until the core wire was exposed. You then inserted a "tap" which was just a probe that touched the central wire. Adding a tap, however, often caused signal reflections around that spot.
All those reflections from new taps meant that adding new machines could disrupt the signal for other connected machines on the cable. Adding new machines to a cable was basically an art since doing anything meant that every pair of machines on the cable had to be rechecked to make sure the reflections didn't mess up communications between those two machines.
Point to point cabling was a welcome and important improvement over this. OTOH, the single cable approach worked better than you'd have thought. CMU's Andrew project started using the single cable approach.
My first use of ethernet was 10base2: it used coax cable similar to TV coax (but 50 ohm impedance rather than 75 ohm). Fortunately, we didn't have to drill into the cable: we had BNC T-pieces. And 50-ohm terminators to put on the T's at the far ends. Maximum length end-to-end was 185 metres.
The whole network would die, either if someone's foot got caught in the cable and pulled it out, or if a dodgy network card would start randomly chattering. Finding the computer with the dodgy NIC was a tricky game of divide-and-conquer, repeatedly dividing the network into halves and putting a terminator at each end.
It was a happy day when we got 10baseT hubs and Cat5/RJ45. Whilst it was still one big collision domain - only one device could transmit at once - the hub could detect a chattering device and block the port.
10base2 was a lot easier to use than the older 10base5. 10base5 cable was a lot thicker and more rigid and usually yellow with black stripes. The black stripes marked where you were allowed to attach another node, by drilling a hole and then pushing spikes into the cable. It was a black art. 10base2 as you say just used BNC tee-pieces and terminators. That was a relief when it arrived. The later 10baseT was a bit of a mixed blessing originally, with the difficult connectors and completely dumb hubs etc that mad fault-finding difficult sometimes.
Yeah. 10baseT was nice for small-office networks, though. You could even run it over "silver satin" – POTS four-wire phone lines – if the lengths weren't too long. That was handy if, say, you were a startup renting some old office space that had already been wired for multiple phone lines; repurpose some of those and voila, every office has an RJ-11 network jack.
In the late '90s someone at famous anime import company AnimEigo (Robert Woodward, maybe?) wrote a web piece1 about hacking both 10base2 and Appletalk into an existing silver-satin network at the company HQ, using one pair of wires for 10base2 and the other for Appletalk, and crimping up a bunch of connector cables that selected the appropriate pair.
1This would be prior to the coining of the obnoxious term "blog" in 1999.
I remember 10base2 - I think I still have a couple of ISA-based network cards (in a box somewhere) that have both 10base2 and 10baseT sockets on them...and a box of BNC T-connectors, some of which have end-terminators on them !
I must be getting old !!
I still have a box (can't recall exactly, but I think there are dozen or so cards in the bulk box) of 3Com Etherlink III 3c509b combo cards gathering dust. Having tried various cards one tends to settle for something that works reliably in most settings.
This post has been deleted by its author
I was at Xerox and used/maintained the 3 MBit ethernet cable. I never noticed much reflection from the tabs but did notice reflections from the cable connectors. The RG11U foam had a characteristic impedance of 75 ohms but the connectors were designed for a 50 ohm cable. The TDR I built could count connectors but barely nodiced the vampire taps. For a really long cable there were problems until I designed a bidirectional repeater which was placed in the middle of the long run.
You could say Robert Metcalfe was at the right place (Xerox PARC) at the right time, but no one would have predicted that Ethernet would become the universal network technology on LANs. The 1980s had many vendors providing LAN technology like Corvus Omninet, SMC ARCnet, 3Com Ethernet, Proteon ProNET 10, IBM PC Network Baseband, IBM PC Network Broadband, IBM Token-Ring, Appletalk, Gateway G-Net, pulse more even forgettable LAN networking schemes from Allen-Bradley and others. That said, the ONE thing that propelled Ethernet into the future was running it over unshielded twisted-pair (UTP) cable and introducing standards-based cabling systems. SynOptics pioneered Ethernet over UTP cable in 1988/89 with LattisNet, which jumped the gun on IEEE 802.3 Ethernet. Still, the concept was proved, and it took off like gangbusters once enough network techs were convinced you could run 10Mbps Ethernet over UTP cabling. The second other thing was EIA/TIA standards for building structured cabling systems, making it easy to design and deploy cabling systems for LANs. Before that, almost every LAN vendor had its cabling system. IBM had its own IBM Cabling System. DEC had its cabling system. The 1990s settled all that, including the demise of Ethernet's primary alternative, Token-Ring networks. Pretty interesting that in the space of 20 years, Ethernet over UTP cable became the winner. Now every mainboard has an embedded Ethernet controller. Congratulations to Robert Metcalfe on receiving his Turing Award as the co-inventor of Ethernet.
The significant expense in data cable wiring is labor, not the cable itself. With many vendors building Ethernet cards, the price became very competitive. Ethernet cards were considerably less complicated to make than Token-Ring cards. That said, Dr. Olaf Solderblum, from Sweden, was busy asserting his patent claim for token-passing networks. He had once worked for IBM. His patent licensing fee paid by Token-Ring card manufacturers contributed to the higher pricing of Token-Ring cards. There were also far fewer Token-Ring card manufacturers than Ethernet card manufacturers. Curiously, Madge Networks in England refused to pay for a Token-Ring license, took Dr. Olaf Solderblum to court in England, and was upheld by the court in its refusal to pay for a license to make Token-Ring cards.
Yup, and the theoretical benefit of bounded wait time for Token Ring (and Token Bus) proved largely irrelevant in practical use, when experiments showed CSMA/CD would converge quickly on a decent fraction of maximum theoretical throughput. Then switching came to Ethernet and people had much smaller collision groups anyway, so it was rarely an issue.
I liked Token Ring, particularly when it too adopted Cat-5 cabling, but I didn't actually miss it when we switched to Ethernet.
Although Ethernet was the big winner in local area networking, there were others who were working on local area networking before the Metcalf paper was published.
I would like to recognize the late Harry Pyle at Datapoint. He started work on ARCnet in 1975 which was commercially deployed in late 1977.
At Datapoint, the joke was that while Ethernet was being advertised as something big coming, Datapoint already had customers for their technology.
Also, the original Ethernet was bus-oriented which required a length of coax with multiple connectors and a terminator at the end. Meanwhile, both Datapoint's ARCnet and IBM's TokenRing were designed as hub and spoke, which is how everything works today.
So, hats off to Robert Metcalf, but lets not forget the other pioneers, as well.
I had the pleasure of meeting Bob at a conference may years ago. Walked up, said hi, started to chat and 30 minutes later the conversation was still going on with him telling stories about 3Com, USR, etc. Genuinely nice guy, not the least bit full of himself. Fair few people can claim an engineering achievement with such reach and longevity. Very happy to see him recognized.
Ah, Token Ring. Fond memories of that.... Well, that's all of them.
The implementation I worked with was Token Ring over STP. (although the wall to machine patches were standard Cat 5 UTP patches...) The place also had a MAU switch/hub in the same rack as the STP based one, probably for historical reasons, or the one or two things that still needed it. (It's been a LONG time since I've been to that place, and I'm quite certain they've ripped it all out at this point.
[makes hex sign]
Token Ring was a nightmare from beginning to end. The shared media version, with the connector panels and bulky, finicky cables was a lot of work for 16 mbits, the interface cards were 3x the cost of an Ethernet card because of licensing fees, and the UTP version radiated like a banshee. A triumph of [IBM] marketing over engineering.
I developed TR cards at Data General and 3Com, and while it gave me a job, I always felt that Ethernet was going to be the winner.
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
