The History of Ethernet

Z34107 tips an article at Ars about the history of ethernet, from its humble beginnings at Xerox PARC in the mid-'70s, to its standardization and broad adoption, to the never-ending quest for higher throughput. Quoting: "It's hard to believe now, but in the early 1980s, 10Mbps Ethernet was very fast. Think about it: is there any other 30-year-old technology still present in current computers? 300 baud modems? 500 ns memory? Daisy wheel printers? But even today, 10Mbps is not an entirely unusable speed, and it's still part of the 10/100/1000Mbps Ethernet interfaces in our computers. Still, by the early 1990s, Ethernet didn't feel as fast as it did a decade earlier. Consider the VAX-11/780, a machine released in 1977 by Digital Equipment Corporation. The 780 comes with some 2MB RAM and runs at 5MHz. Its speed is almost exactly one MIPS and it executes 1757 dhrystones per second. (Dhrystone is a CPU benchmark developed in 1984; the name is a play on the even older Whetstone benchmark.) A current Intel i7 machine may run at 3GHz and have 3GB RAM, executing nearly 17 million dhrystones per second. If network speeds had increased as fast as processor speeds, the i7 would today at least have a 10Gbps network interface, and perhaps a 100Gbps one."

Comparing high end to low end

[milgr]

In the 1980s, ethernet tended to be over Thinnet or Thicknet. I seem to recall speeds of 1-3Mbps over those technologies. Twisted pair came out somewhere around 1990 at 10Mbps.

Today I mostly use 1Gps, but deal with servers that are 10G.40G and 100G will be standard in datacenters in a few years.

The blurb indicates that Ethernet is the only technology that we are using from 30 years ago. Back then all the machines I used had Memory, cpus, displays, and keyboards. The particualr technology changed - just like Ethernet technology's changes.

Ethernet, the early days

[Animats]

Now this really dates me. But in 1975, I got a tour of Xerox PARC when I was taking a summer course in computer architecture at UC Santa Cruz. Alan Kay showed us some of the early Alto machines. They were still having trouble getting a smooth phosphor coating on the custom-made page-sized CRTs. We saw the PARC 3mb/s Ethernet, which Kay described as "an Alohanet with a captive ether," the first networked file server, and the first networked laser printer. It was clear this was the future, if the price could come down by about a factor of 10. Kay was hoping that some day a workstation might cost as little as a grand piano.

At Ford Aerospace, I was responsible for putting in the first Ethernet, around 1981. It was mostly "thick Ethernet" at 10mb/s. Ethernet cables weren't standard items, but Ford Aerospace routinely built cables for satellite ground stations, so we had the appropriate cables made up and pulled through the telephone ducts run through the building's concrete floors. I checked out a time-domain reflectometer from the measurement equipment pool and took a look at the cable. Cables ended in PL-239 coax connectors, and sections were joined with a barrel. The Ethernet tranceivers had SO-239 connectors on both ends, so the cable went through them. We used a vampire tap once or twice, but it didn't work out as well. The TDR showed a transceiver as generating almost no reflections. But bending the cable tighter than a 1' radius caused a noticeable impedance mismatch.

We were bothered that coax Ethernet wasn't a balanced system. There's a DC component to the signal, which means you can't use decoupling capacitors between sections to get rid of hum. We spent time on grounding issues and looked at the cable signal with scopes a lot. Repeaters were very expensive then, and we were trying to avoid them.

The network interfaces were mostly 3Com boards. Our original network consisted of a PDP 11/70, a PDP 11/45, a VAX 11/780, and a PDP 11/34 used as a gateway to a 9600 baud leased line "backbone link" to Ford HQ in Dearborn MI. We later added four Sun 2 workstations and a Sun server. Everything ran TCP/IP. Ford HQ had a similar link to Ford Aerospace in Colorado Springs,which had an ARPANET IMP. So we could get to the ARPANET over a 9600 baud shared backbone. We could FTP files instead of mailing tapes! I used to Telnet into Stanford's machines over that link.

I did a lot of work on 3COM's TCP/IP implementation, which originally was totally incapable of coping with a mix of speeds in the network. That's why I have those RFCs on network congestion with my name on them. This was before telephone de-regulation, and that 9600 baud leased line was expensive.

The article mentions that "There used to be a lot of fear, uncertainty, and doubt surrounding the performance impact of collisions." There was a period around 1984-1990 when coax Ethernet performance in practice was much worse than theory predicted. The problem was finally figured out by Wes Irish at Xerox PARC. [google.com] It turns out that the defective design of a SEEQ Ethernet interface chip was causing the problem. As the state machine of the chip transitioned at the end of receiving a packet, there was a period of a few nanoseconds when the chip momentarily turned on the transmitter power, jamming the coax for a few nanoseconds. This reset the "quiet time" timer on all the other stations on the cable, causing them to ignore any following packet for several microseconds, after which they dropped back to the proper "look for sync" state. Back-to-back packets thus lost the second packet, which caused retransmissions and killed performance, but didn't show up as a "collision" to the controll