Transcription

This simple cable, unshielded twisted pair wire, terminated in the RJ45 connector, carrying Ethernet signals. But that's not at all how Ethernet started. So how did we get from this, to this? Today on the Serial Port, we're going to be exploring the hidden past of twisted pair Ethernet. From its beginnings in the early 80s, its near derailment from two giants of Ethernet, and how it ultimately prevailed to become the network of the world. So come with us on a journey that spans nearly a decade, as we uncover, twist by twist, the story of twisted pair. It's easy to forget in today's world where Ethernet has been the standard of local area networks for what seems like eons, but its success was never guaranteed. As John Reed demonstrated in our last video, Ethernet in its 10BASE5 form was difficult to work with, to say the least. Installing the large, inflexible coax wiring was expensive and difficult to troubleshoot. Just as soon as it appeared, there were calls for an easier and cheaper solution, especially as local area networks began to enter office environments. One answer arrived in the form of a new implementation of Ethernet called 10BASE2. Finalized in 1985, 10BASE2 employed the use of a smaller diameter coax cable, which allowed cables to interface directly with NICs through VNC connectors, eliminating the need for an external AUI transceiver. Because of its thinner cabling, it became known as ThinNet as opposed to ThickNet 10BASE5, and the lower cost also brought about the moniker of CheaperNet, as 10BASE2 quickly took the marketplace by storm. It shared many of the same issues as 10BASE5, though, in distances between nodes were now capped at 185 meters compared to the 500 meter limit of 10BASE5, and like its older, more expensive cousin, because of its bus topology, it was easy to bring the entire network down. In 1983, Sally Ride became the first American woman in space. The first mobile phone arrived, and the official birth of the internet took place on January 1st. But unbeknownst to almost everyone on the planet at the time, a project was being formed that would coalesce into one of the most important advances in networking history. And indirectly, it all sort of started with a chip. Meet the Intel 82586. Released in October of 1982, this was a co-processor that supported the original DIX version of Ethernet. It was widely used across Ethernet devices, and drastically reduced the cost and complexity of developing those products. And on the heels of its success, Robert Gallin, a product manager at Intel, was tasked with finding even more applications for an upcoming successor to the chip. Gallin eventually got in touch with four different companies, all looking to craft their own LANs. National Cash Register, or NCR, Wang Laboratories, AT&T, and Tandem Computers. AT&T had been hard at work developing their 3B2 microcomputer, and they had an interest in being able to cost-effectively network the systems together, as the intended environment for these systems was the business office. That's when Tim Rock of AT&T Information Systems in Holmdel, New Jersey, began researching possible solutions. AT&T had just been dealt a massive blow to their status as a monopoly, with the United States v AT&T decision enacted on January 8th of 1982. They were suddenly stripped of ownership of a huge number of assets, namely, all of the existing telephone cabling installed in thousands and thousands of commercial buildings across the United States. Snaking through virtually every office building in the country, were bundles of telephone wire made up of what was called Unshielded Twisted Pair, or UTP. This wiring had carried the signals of billions of telephone calls for decades, but Rock realized it held potential for another application, computer networking. The cable held 25 pairs of twisted pair wire, in a simple unshielded bundle. Because of the way multi-line office telephones worked back then, a full 25 pair cable was typically run to each office in a building, but not all pairs were used in most cases. Almost always some unused pairs in there, and if you could do local area networking with a couple of pairs, you know, that'd be wonderful. Gallon then got in touch with this guy, Richard Bennett of Tandem Computers. The Tandem Computer Corporation, located in Cupertino, California, was a startup from ex-Hewlett Packard employees that focused on fault-tolerant systems for applications like banks and stock exchanges. And if you know anything about the history of mainframe computers back in those days, the process of upgrading from one version of an IBM 360 or 370 to the next one up, it would take months to do an upgrade, in short. With the Tandem, you know, you just like plugged in another CPU module, and you're good to go. Bennett was researching networking solutions for Tandem systems when he got in touch with Gallon and Rock. And we came across the idea that, well, actually what it should be is, it should have kind of a topology like Ethernet, where it doesn't really care how many devices, you've got like one interconnection point, we don't care how many devices are on it, right? And so we were interested in it, and we were in regular contact with Intel, because Tandem was sort of at the leading edge of a lot of networking at the time. All five manufacturers realized that they were after the same thing, a low-cost, easy-to-install local area network. They decided to join forces and bring their idea to IEEE 802.3, the group responsible for developing Ethernet standards. And the IEEE 802.3 was interested in the idea that we pitched to them on basically Valentine's Day of 1984, as I recall, that we wanted to solve the cost problem with Ethernet. And in our estimation, the cost problem came down mostly to the installation of wires. Ethernet was already in the works at that point, which did a whole lot to reduce the cost of the physical cable, but the labor component of it was still there, and in most cases even as expensive as classic Ethernet 50-ohm cable was, the labor was more expensive. And so by essentially using unused pairs that were in the telephone bundles in office buildings, we essentially reduced the cost of wiring to zero, and you know, it doesn't get any cheaper than that. Well, I'll hold my beer while I go get Wi-Fi. During this time, each phone in an office building typically had a 25-pair or 4-pair telephone cable that terminated at a central distribution unit in a wiring closet. In other words, the phones were spokes connecting at a central hub, but utilizing the existing telephone cabling for LAN meant using this same spoken hub concept and doing away with the bus topology of the existing Ethernet standards. Each node would instead connect in a star topology using a central device that would interconnect each node. This however, was not a popular idea at the time. You know, when we pitched this idea to 802.3, Ron Crane, who was one of the founders of 3Com, he wanted to replace our hub and spoke topology with a bus, and so he had kind of figured out some way that he could get, you know, Ethernet-like behavior out of a single twisted pair bus, but it didn't really go anywhere. And to physically interface everything together, the group wanted to use the same connector that many office phones had begun to use. The Registered Jack 45, or RJ45. This connector provided space for 8 contacts, plenty of space to carry both phone and LAN data. And to be accurate, the connector used for twisted pair Ethernet isn't technically an RJ45, though they are almost identical. The true RJ45 is keyed, with the slot shown here on the right, whereas the variety everyone is familiar with lacks that feature. It's kind of funny, a few months, there's a subreddit called Home Networking, and somebody expressed some rage on there. Whoever that was that decided to use RJ45 for Ethernet, because it was the worst choice ever. And, you know, the man needed to be, he wanted to punch him out. And I said, raise my hand, well, I'll take the blame for that. And it wasn't my personal decision. I mean, I kind of was the tie-breaking vote, I think, on that. So I said, hey, you can blame me, sorry, but, you know, it seemed like a good idea at the time. Their idea had enough merit, though, that IEEE agreed to form a task force for Ethernet over unshielded twisted pair, with Gallon acting as chairman and Bennett as vice chair. And then once we were sanctioned as a task force, then, you know, it was on us to come up with technical proposals and share those with the larger group of 802.3 and, you know, try to win their support. With twisted pair finally on a standards track, AT&T had already started to develop products for the new technology. It was named StarLAN, an obvious reference to its star topology design. It used two sets of twisted pair wire, one pair for receiving and the other for transmitting. Systems that had a StarLAN NAU or network access unit, in other words a NIC, could be either daisy-chained together or individually connected to a central element known as the NEU or network extension unit, which is equivalent to what we would call a hub today. And StarLAN speed was capped at a comparatively slow 1 megabit per second, thanks to both limitations of the Intel controller chips and how StarLAN dealt with errors introduced by poor cabling. Several other companies began showing interest in the technology, and as computers began entering offices at higher rates, the benefits of using twisted pair became more salient. And this was a time when the landlords began heating up, as IBM's token ring had recently hit the scene. No longer considered the protocol of the future, Ethernet was predicted to have to share space with several others, so the success of twisted pair began to carry more weight. Of course, there were already other LANs that used unshielded twisted pair wiring, like Corvus OmniNet, but this system used a bus topology and was otherwise proprietary. And there was a third-party system called PhoneNet for AppleTalk networks. But eventually the new IEEE standard was finally completed and published on June 11th of 1987. When the company coined 1BASE5, it used the StarLAN scheme of 1 megabit per second, but was only half-duplex, meaning it could not send and receive data simultaneously, just like the coax versions of Ethernet. And so we were just kind of scared that if we departed too far from classic Ethernet, they would freak out and wouldn't, you know, wouldn't let us draft the standard. So we wanted to keep it as much like classic Ethernet as possible, but open the door to, you know, sort of further enhancements around the hub-and-spoke architecture, which is, you know, that's what happened. I mean, StarLAN was kind of a throwaway standard, but it's almost better viewed as a proof of concept than a legitimate standard. While maybe not a huge commercial success, StarLAN did show that Ethernet over twisted pair was possible. There was skepticism, though, whether it could ever achieve higher data rates, but that's when everything changed. Hello, is this Pat? Yes. Hey, this is Ben Grubbs from the Serial Port, how are you? He got in touch with Pat Thaler, who had worked on StarLAN as an engineer while at Hewlett-Packard, but she would later become instrumental in the success of twisted pair Ethernet. I was working on interface cards at Hewlett-Packard. Our group started working with some of the semiconductor vendors, Intel and AMD mainly, on the first integrated Ethernet chips, making sure that they worked, that they behaved properly and met the spec and all that. I volunteered to go to the standards meetings because it was kind of annoying to work one remove and to not be directly in there involved in setting the specs that I was making sure were met. Thaler was thrown into the gauntlet of standards development by having to face off with deck engineers over the multiport repeater standard that was in development. At this time, Deck had their own multiport repeater, but didn't want to openly share their design and technology with the standards body. But getting the repeater standard finished was critical for Ethernet success against Token Ring. My boss, who was the chair of 802.3 at the time, he talked to the standards management at Deck and said, you know, this is hurting us in comparison to Token Ring because Token Ring guys are saying, well, they can't finish the standard, they can't fix this repeater. So he convinced them that the real enemy out there was Token Ring and we needed to fix the standard. So then I was supposed to go back east to Massachusetts to meet them, get them to agree on what they wanted in the standard so that we could finish it. They didn't actually 100% commit to the meeting. They kind of said, well, yeah, we'll probably meet with you. I flew out there and they did meet with me and it was like five engineers, six engineers and a manager in the room with me. They were pretty arrogant guys, but because they felt they'd developed it all. So we got down to it and got an agreement on what would go into the standard and go back to the next meeting and get the final thing drafted and finish it up. Baylor then worked on the OneBase 5 Starland standard, but in her mind, there was still untapped potential. We finished the one megabit standard. We at HP felt like, you know, there was a lot of slack left by that standard, that we weren't really pushing things. The standard was actually designed so that you could use existing components and build it out of small scale integrated components, SSI, and put it together as a circuit board. That approach meant that needed a lot of margin. And we started looking at, could you go faster? Could you make it go 10 megabit if you really designed components specifically for this? Could the cable do it? We came to the decision that it could. Power environments are constantly bombarded by electromagnetic interference, or what's called EMI, which can show up as noise in a transmitted signal. If you're trying to transmit digital data, the recipient needs to be able to distinguish a one from a zero reliably, but with lots of noise, this becomes increasingly difficult. 10Base 5 and 10Base 2 solve this issue by using a heavily shielded coaxial cable, which helps prevent noise from affecting the signal. The Ethernet signal is transmitted as a voltage potential between the center conductor and the shield, but twisted pair was a different animal. Yeah, so twisted pair is what's called a balanced medium. You know, if you have a coax or something like that, that's kind of unbalanced. You've got a shield wire and you've got an internal wire running through it. If you've got a twisted pair, you've got two wires and they are continually kind of rotating in space, twisted around each other. So because they're very close to each other and because they're constantly switching positions, any noise one wire in the pair picks up is pretty much the same as the noise the other wire in the pair picks up. And when you're looking at the signal to decode it, you're not looking at how the signal compares to ground. You're looking at how the signal on those two wires compares, the differential, in other words, between the two wires. However, an added challenge with twisted pair cabling is EMI from other pairs in the same bundle, which is called crosstalk. Crosstalk occurs when a signal from one twisted pair induces a voltage in another. This happens because when current flows through a conductor, an electromagnetic field is generated which can influence or couple to any nearby conductors. And in a 25 pair cable, there could end up being several different signals in adjacent pairs to contend with. The higher speed you go, the more you have to worry about crosstalk, the better crosstalk couples. So there was a lot of testing work to say, this existing cable built the way it is, what's the actual performance when you put higher frequency signals on it? Does it still behave linearly? How does the noise couple? What's the noise in a 25 pair bundle? And it turns out in a 25 pair bundle, you're kind of next to like about six pairs and then those pairs shield you from the other pairs. We found that, yeah, there was crosstalk, but it kind of maxed out at a certain point. And we could deal with that. We could deal with tolerating that amount of crosstalk. With the capability of 10 megabits per second within reach, Thaler and her colleagues were ready for the next step. And we approached AT&T to work with them on it and a couple of other companies. And we started kind of gathering some support. And then I went to IEEE 802.3 and made the proposal that we work on standardizing 10 megabit over twisted pair. And at Thaler's request, a study group was formed in July of 1987 to work on a new standard, one that would alter the course of Ethernet for decades to come. It was called 10BASE-T. With the 10BASE-T study group now formed, they first and foremost needed to characterize the real world performance of unshielded twisted pair at higher speeds. It was one thing to show 10 megabits per second was possible in the lab, but the real world was a different story. And there were probably at this point about five or six different companies involved doing testing. NCR did a lot of the testing, for instance, finding out, okay, we've defined what the crosstalk is from kind of our own signals linking into us. But in these buildings, what other noise do you pick up? How bad is that noise? And we went into a lot of existing buildings and would put monitors on and kind of capture the highest noise spikes that happened during a couple of days. The Dayton YMCA, which NCR tested, that had some of the worst noise. I remember that that was kind of our worst noise candidate, Dayton YMCA. And development had already been ongoing even prior to the study group. One member of the group, Synoptics, announced their own version of 10 megabit UTP Ethernet coined LatticeNet in January of 1987. And HP announced the Starland 10 system in August of the same year. But as the study group refined their data and got closer to a proposal, a rift began to form inside the group. 3Com and DEC, both participants in the study group, had already invested heavily in 10Base5 and 10Base2 technology. 3Com alone had an installed base of more than 400,000 coaxial transceivers at the time, and were shipping over 20,000 more every single month. Bob Metcalf, the founder of 3Com, had been vocally skeptical of the first Starland standard. And he, along with Ron Crane, were not fans of star topologies with a central hub element. I mean, it essentially required admitting that the Ethernet itself was the mistake. And it was mistaken with very good intentions. I mean, like what Metcalf was worried about back in those days was that he didn't want to have electronics at the center of the network because he believed that if you did, the electronics would be a bottleneck. And so that's why he really believed this passive backbone thing was the way to go, because there's no way it could be a bottleneck, because it didn't really care what you were doing to it. But in fact, he kind of missed the, like, the key point is that the switch, in order to avoid being a bottleneck, it doesn't need to have like more capacity than all the entire population of devices that are on it at the same time. It just needs to be as fast as the interface on each individual, on an individual device. I mean, if it's dedicating a port to each device, there is no bottleneck problem. I mean, maybe in, you know, in the interior of it, but, you know, it's really easy to make electronics go fast. The group's trajectory eventually ended up creating two opposing camps of manufacturers. One, made up of the behemoths of three common deck, wanted to use a hybrid solution where a single twisted pair cable would be used to connect a workstation or to bridge to another coaxial network, preserving their installed base of coax devices. The other, known as the AUI group, was made up of AT&T, HP, Synoptics, and others that wanted to build a new transceiver and use the central hub model that the previous Starland standard had used, creating an end-to-end chain of unshielded twisted pair. This resulted in a six-month stalemate as the two groups battled for their proposals, and it was Thaler's task to try to reconcile their differences. And to add even more pressure to the situation, the media attention this generated was constant. Yeah, it was, it started almost right from the start. It was, 10BaseT got a lot of attention. I mean, there were, I had reporters calling every week or two, and sometimes several, because there'd be kind of concern about, oh, are they ever going to finish this? You know, there'd always be, oh, there, there, there's this deadlock in the committee and they can't get consensus on which proposal to go forward with. But ultimately on March 17th of 1988 in San Diego, California, the AUI group prevailed in a vote with three common deck withdrawing their proposals, signaling the first death knell of Ethernet coax. The fervor for Twisted Pair only grew as the products from Synoptics, HP, and AT&T had already started to hit the market. But 3Com, the company once on the leading edge of Ethernet, said they had no plans on releasing products supporting the winning proposal, as they found themselves on the back foot in the Ethernet world for the first time. The anticipation was so hot that the IEEE organization had to issue a rare statement in November of 1989 that any products claiming 10BASE-T compatibility were technically not compatible until the standard was actually released. But after ironing out the final details, the 10BASE-T standard was finally published on September 28th, 1990, the culmination of a near decade's worth of work to bring Ethernet over Twisted Pair. 10BASE-T products flooded the marketplace and rapidly expanded into all areas of networking. One of my children, my oldest, was just starting to look at colleges and we'd go to colleges we were looking at and they'd say, you know, that they were putting it in the dorm rooms. They had one connection now per dorm room. They were hoping to get to having one per, one per pillow. It really exploded. People started putting it in their houses. You know, it was, it was a coinciding of finishing the standard and then the explosion of personal computing and home computers and desktop computers. To celebrate the incredible achievement of 10BASE-T, we've procured the first two pioneers of 10 megabit unshielded Twisted Pair for inclusion in the Serial Port Museum. The Synoptics LatticeNet Concentrator and the AT&T Starland 10 Hub. Starland 10 is significant as 10BASE-T signaling is largely based on it, but Starland lacks what's called Link Beat, which is how 10BASE-T determines the link status. LatticeNet also lacks Link Beat and instead uses a DC signal on the receive pair to indicate link status. While they aren't strictly 10BASE-T compatible, we are inspired to see if we can get these working together, so please subscribe to the channel if you haven't, if you want to keep up with our progress. The story of Twisted Pair is one of struggle and persistence. Though coax would stick around for some time, 10BASE-T was the first standard that introduced Ethernet that is recognizable to us today. 10BASE-T marked a turning point where local area networks could be installed affordably and quickly, setting the stage for the explosive growth of home networking. And Twisted Pair grew to unimaginable heights, with the latest standard at an eye-watering 10 gigabits per second. And it's thanks to the determination of people behind the scenes, like Bob Gallen, Tim Rock, Richard Bennett, Pat Thaler, and so many others who worked tirelessly to advance networking, and in turn, the world. Thanks for watching The Serial Port, and we'll see you next time.