Seeing as I had just posted a Fast Ethernet flashback, I suppose it would be a good time to go further and look back another decade to the “original” Ethernet.
Prior to the development of an over-arching networking standard, networks were very much proprietary monoliths. The physical layer, the protocols, the software were all pretty much bundled and system specific, meaning that only computers of a certain vendor would “talk together”. Such networking systems often had lower speed, and varying levels of robustness, for example, the RS-422 based LocalTalk network which could be run over phone lines between Apple computers operating at 230.4kbit/s maximum, which was fairly common when I was in primary school; or ARCNET operating at 2.5Mbit/s which was mostly used with IBM and Tandy machines.
The 802 project to standardize LANs was important to allow equipment of any vendor to co-exist, interoperate and exchange data between one another. The success of the Carrier Sense Multiple Access / Collision Detection (CSMA/CD) format Ethernet which we still use today was far from assured. At the time that Ethernet was being proposed, it was a battleground between the competing Token Ring standard backed by IBM, and a variant of it named Token Bus supported by General Motors, and even after it was standardized and published in 1985 (30 years ago), it was still under attack from campaigns to discredit Ethernet’s performance, speed increases in the competing Token Ring network to 16Mbit/s and proprietary standards which emerged afterward such as LattisNet and StarLAN-10 which competed against 10BASE-T.
Some of these were solely rumours, which were unfounded. Some that I had heard were so strange as to be unbelievable – for example, that unshielded twisted pair could not ever work because the signals will just leak into thin air. Others were firmly placed against Ethernet’s un-coordinated transmit mechanism, claiming that under high loading, collisions were so frequent as to cause the network to collapse entirely. As it turns out, such scenarios were extremely unlikely solely due to the fact that the “window” in which a collision can occur is so short, and the cards operate so quickly, that collisions are rare. From memory, a vendor claimed that measurements on their corporate network revealed a collision rate of about two per month, with no ill effect as the lost packet was immediately retransmitted.
It was fortunate, then, that none of the competing options gained enough traction against an industry standard to bring it down, and neither did the rumours and misinformation, as this set up the future for Fast Ethernet and affordable, reliable, and standardized networking for all.
The original Ethernet was a shared medium system, introduced as 10BASE5 utilizing thick RG-8X copper coaxial cable runs, which were stiff, expensive, but also low-loss, resulting in a 500m range. These cables were terminated at both ends with a 50 ohm resistance, and one end was grounded. Up to 100 nodes can be added to a single live network segment using a vampire tap at 2.5m intervals to avoid signal reflection, and transcievers would be attached to the cable, connected to the NIC via an AUI connection (15-pin). This was a relatively expensive and clunky solution requiring specialized installation, and so a cheaper alternative became common.
Only a few years later, a “cheapernet” using thin RG-58 coax was born, known as 10BASE2. This cut down the range to 185m, but allowed for up to 30 nodes in a network using standardized BNC connector hardware. Nodes cannot be added to a network without disrupting the network, due to the need to open up the bus. A comically short network cable segment is shown below with BNC T-pieces, and 50-ohm terminators on each end.
Due to the physical “bus” nature of the network, it is basically connected in a daisy chain. Troubleshooting such networks could be particularly difficult if there is an intermittent fault in the middle, as it is often an “all or nothing” result. These networks were relatively cheap as there was no significant network infrastructure to invest in, unless you reached the maximum number of nodes or distance, where repeaters or gateways would be necessary. This was quite popular for many years, with the thin-net transceiver hardware being integrated onto many networking cards instead of externally connected via AUI. In fact, in UNSW’s Electrical Engineering building, there are still some BNC socket pairs used for 10BASE2 in some rooms. It was also very commonly deployed during my primary school days (late 1990’s, early 2000’s) for networking library computers for catalog searches, loans and returns administration.
If you had a laptop, you didn’t have to miss out on the BNC thin-net action as seen in the break-out leads above for both 10BASE-T and 10BASE2, as well as telephone modem.
The alternative medium of twisted pair cabling also came to the fore, along with a physically star connected network, with the advent of 1BASE5 which was commercially unsuccessful 1Mbit/s over twisted pair standard along with StarLAN, which paved the way for 10BASE-T to be ratified in 1990 several years after LattisNet and StarLAN-10 proprietary solutions had demonstrated successful 10Mbit/s operation. The intention behind the 10BASE-T standard was to enable operation with existing premises telephone cabling, down to CAT-3 for more inexpensive installation. This idea of twisted pair cabling has now become the most common media which Ethernet is run over to the present day.
All cables from the end nodes lead to a central “hub” which makes the network physically cabled to resemble a star, but as most hubs/concentrators were not intelligent, the signals received on one port were “echoed” out on all other ports making the network topologically resemble a bus network. Switched Ethernet was to come several years later to improve efficiency by only sending packets where they needed to go, and allowing several exchanges to happen in parallel as long as they didn’t involve the same pairs of nodes, which allowed the network to take a more star-like topology.
Designing larger Ethernet networks were particularly interesting due to the idiosyncrasies of repeater hubs which “consumed” bits of the preamble as they were locking onto the data stream. Stacking hubs together would potentially cause the packets from the end nodes to have preambles which were too short for the destination NIC to handle, or even loss of part of the packet itself. As a result, there were strict rules about how many hub levels were allowed, and different vendors had different ideas – one was the 5-4-3 rule, and the other was more restrictive, only allowing two repeaters with a dedicated link between the two. This is no longer a problem with more sophisticated electronics which can store and forward (i.e. regenerate) transmissions in their entirety.
As a person of the younger generation, by the time I was born, many of the issues had already subsided and by the time I came to network my computers, Fast Ethernet was just becoming affordable enough to consider. I honestly didn’t operate 10Mbit/s Ethernet for very long (both thin-net and twisted pair using discarded ex-corporate equipment), and my collection of equipment is relatively limited late pulls mostly originating around disposed gear. I remember fondly of wishing there was a faster way to transfer files than through a Null Modem Cable (or even a Laplink Parallel Port cable), but network cards were still somewhat expensive and networking support was not baked in, or easy to configure. TCP/IP was often run with other protocols such as IPX/SPX, and numerous network operating systems were common. These required more in-depth configuration, and were more often found in businesses rather than the home. But despite all of this, I remained aware through computing magazines, encyclopedic articles and from IPT studies of the climate under which Ethernet was born and its enduring legacy.
So lets celebrate 30 years of Ethernet by looking at the 10Mbit/s cards in my collection, which are mostly later cards.
Racal Interlan NI5210B 625-0173-00
This is the oldest card in my collection so far, already partially covered in the article about the system it was pulled from. It is an 8-bit ISA card, with many discrete components well packed in and jumper configuration. It has an Intel P82586 Ethernet Co-Processor socketed at the top, but other than that, most of the other chips appear to be different CMOS logic, RAM, and an unknown transciever possibly. There is a large DC to DC converter module from Valor, which is quite common on 10Mbit/s cards, to provide the required voltages for the line drivers. A neon bulb can be seen acting as surge suppressor as well.
The card is Made in Malaysia and can be dated to Week 43 of 1991 according to the PCB.
As an earlier card, prior to the popularity of 10BASE-T, this features integrated “thin net” BNC connection, and an AUI connector for thick-net connection. A switch is provided on the rear panel to switch between the two network types. Curiously, no troubleshooting LEDs are provided.
Racal EtherBlaster T2 620-0381-01
The second-oldest card in the collection is also a Racal Datacom Inc. product, being a 16-bit ISA card. It can be dated back to Week 32 of 1993 by a Valor IC on the card, and is based around an integrated AMD PCnet-ISA AM79C960KC chipset, a later version of the very widely compatible Lance design. Just one configuration jumper can be seen, and so this card is likely to be ISA PnP. The card has baked in support for 10BASE2 and 10BASE-T but not 10BASE5. Unusually, this has an Option ROM installed – a Lanworks Bootware NE2100 boot ROM v1.12, possibly for network booting of Windows NT.
One green indicator LED is visible on the back, which is useful or troubleshooting.
Intel EtherExpress PRO/10+ LAN Adapter
Of course, Intel was in the game as well, and this particular adapter was one of my favourites for the 16-bit ISA bus. It is based around Intel’s FA82595TX integrated chipset, and is fitted with Intel’s own Option ROM, and appears to be dated about Week 19 to 34 of 1994 by the components fitted. By then, it seems progress in switching converters may have improved and clunky DC to DC modules seem to have miniaturized.
The PCB is dated to Week 32 of 1994.
The most likeable part of the card was relatively good OS support, and all medium support – 10BASE2, AUI for 10BASE5 and 10BASE-T are all standard. Two LEDs are provided for link and activity as well, although at this time, 10Mbit/s was on borrowed time as Fast Ethernet was to be standardized very shortly, and become widely adopted in 3-5 years.
SMC SMC8432T EtherPower 10Mbps
Here, we see the first example of a PCI network card, from SMC, which was a timely effort given PCI was introduced in 1992. This chip seems to date around Week 48 of 1995, and is built around a Digital 21041 with a socket for Option ROM. The PCB itself seems to be dated Week 10 of 1996, and the label implies Week 14 of 1996 which might reflect the difficulty of selling a PCI adapter when many machines still had ISA buses as standard. Regardless, this design is highly integrated, with many unpopulated areas. It seems the same PCB design could be used for an “all medium” card should the appropriate parts be populated. There is no wake-on-lan socket (of course) as the concept had not been invented yet.
Rather surprisingly, this unit features integrated LEDs in the jacks – earlier than I had expected.
SMC 668-004 EtherPower 10Mbps
A smaller, more modern card to the one above was also seen.This one is dated Week 4 of 1997 and features a substantially cut-down version of the PCB above, with only thin-net support footprints which remain unpopulated. The chipset itself appears to be identical, which explains why SMC had an easy time of driver support, as many of their cards may have used the same chipset.
It has a similar external branding, with integrated LED jacks. However, a second version was also spotted with new markings on the chip, which may reflect a later version of the chip.
The PCB and backplate are otherwise identical, and are dated to Week 42 of 1997. It is refreshing to see all of the SMC adapters were Made in the USA, in a time back when not everything was Made in China. Maybe that reflects the real SMC prior to the acquisition by Accton and explains how the brand slowly lost its value.
By this time, cheap and generic was starting to infiltrate the 10Mbit/s market as well, and these cards may have been a “cost conscious” choice for those who could not justify shelling out for Fast Ethernet or wished to continue working with 10BASE2 networks. The clunky DC-to-DC converter module makes an appearance, along with the option ROM socket and a cost-reduced “four wiper” RJ45 jack.
The components on the board typically date it to Week 11 to 21 of 1997, with the PCB claiming Week 17, and the warranty label indicating it was sold by June/July of 1997.
Two 3mm LEDs provide activity and link indication on the rear. The chipset itself is …
… remarked by D-Link to disguise its source. But it’s likely to be a fairly basic remarked Realtek RTL8029 possibly.
This card is relatively late in the Ethernet game, dated Week 2 of 1998 by the PCB and roughly Week 50, 1997 to Week 6 of 1998 by the components. It is made by Accton, a Taiwanese manufacturer which manufactures for many other brands and generics. This card is based around the Realtek RTL8029AS and makes no attempts at hiding the fact. It provides an Option ROM socket and the 4-wiper cost-reduced RJ45 jack and 10BASE2 BNC connector.
Two 3mm LEDs are provided at the rear for indication of activity and link.
By now, nobody is really taking 10-Mbit/s Ethernet seriously anymore, and these cards were put out mainly for compatibility reasons, should people still need to replace cards in older machines working with thin-net. As a result, the components they use become substantially similar, and relatively unexciting and cheap. It is likely the serious vendors had moved onto more sophisticated Fast Ethernet equipment seeing as there was no money to be made in a rapidly cost-reducing market. This one is dated around Week 7 to Week 14 of 2000 by components and Week 13 of 2000 by PCB.
It’s all very plain Jane generic, Made in China equipment. By this time, Fast Ethernet was beginning to become affordable to the vast majority, and BNC cabling was expensive compared to UTP. Increasing bandwidth requirements and the allure of switched as opposed to “repeated” networks very quickly closed the door on the 10Mbit/s Ethernet era.
The original Ethernet was the proving ground where the battle between Token Ring and Ethernet fell in Ethernet’s favour. It was also the ground where we were introduced to UTP cabling, and the beginning of affordable networking. While still relatively uncommon around the home at the time, it was a commodity in most businesses and was vital to their networked operation for file and print sharing. In the end, Fast Ethernet swiftly displaced Ethernet over a period of about 5 years (from about 1997 – 2003) and UTP cabling and star-connected networking has become the primary networking topology in use today.
On another note, I think it is relatively exciting that I can still plug a 10Mbit/s 10BASE-T card (a 25 year old standard) into most network switches made today and still obtain a perfectly serviceable, albeit slow, connection. The enduring power of 10BASE-T seems to lie in the fact that multi-standard transceivers have a very small incremental cost, and for smaller embedded systems which do not need the bandwidth, the simplicity of implementing such a connection can make it the most cost effective route even today.