foone Profile picture
Oct 14, 2020 77 tweets 15 min read
Where's my bnc cables, it's time to build a LAN
It's the 3com EtherLink II!
I'd probably have to dig up an old networking book to figure out 10base2/thinnet. it seems like thing kind of thing that isn't going to have a ton of info online for it.
although really it's not super complicated. The tl;dr is that you need BNC-T connectors, BNC between each PC, and terminators on each end.
you don't have hubs or switches like how 10BASE-T and derivatives have, you just wire it as a big long line of computers.
which has the obvious downside that if the cables from one computer get messed up, you no longer have A network, you have two separate networks for each half
T-connectors are those little T-shaped things.
The terminators are one-sided things you put at either end.
They stop the signal from bouncing off the end of the cable and causing interference.
10Base2 (this kind of networking) was popular early on because it was relatively cheap. Network cards were expensive, sure, but the amount of additional hardware you needed was minimal, and scaled very simply with how many nodes you had online.
this IS Ethernet, btw, just like the ubiquitous kind used today.
Ethernet is the link-level protocol, and it can be carried over multiple physical layer protocols.
The cat5-cable kind we use today is some variation on the 10Base-T sort, T for "twisted pair".
so named because internally it's an unshielded cable composed of multiple twisted pairs of cable.
twisted pair cabling is used because each pair forms a balanced circuit, and any electrical noise induced into it will be roughly identical in both wires.
Then the receiver can subtract out the common interference, getting only the data.
and it turns out doing it this way works well enough for most conditions and is cheaper and easier to work with than shielding the cable.
(Although you can always shield the cable afterwards, and get the best of both worlds)
this is compared with the 10base2 or thinnet shown in my first post, which uses coaxial cable.
This is the kind of cable used for TV connections, where there's a center cable, an insulating sheath, then a woven mesh of another connector, then possibly more shielding after that
This is a more complex way to prevent interference on the internal connector, and while it works very well, it results in cables that are more expensive and stiffer, making them hard to route through desks, walls, and under-floor tiles
Ethernet started with neither of these, it started with thicknet.
This was an even thicker cable than standard coax, with the weirdness that you just had one wire for the entire network.
And you might say "foone, isn't that how 10base2 works, where you design it as a line?"
but no, that's a set of cable segments which act as one big wire, with connections between them.
10base5 actually WAS one wire, with no breaks in it. And yet you could have more than 2 computers. How?
The answer is "vampire taps".
You attached to the middle of the cable, without cutting it.
Vampire taps are these fun little devices which clamped onto the cable and had a big spike in the middle, which'd stab right through the cable's insulation and shielding and make contact with the conductor inside.
The connectors to the cable ended up being big and bulking because of this, as they had to make a good connection and couldn't risk disconnecting as they were moved around.
This also made it annoyingly hard to re-wire your network, as you might expect.
This sort of connection is why you often see those big DB-15 connectors on old network cards, like this one.
that's Attachment Unit Interface, it's designed to let you connect onto the vampire tap. They didn't just build a connection right into the card, because the physical layout of that would be almost impossible.
So instead you'd have a Medium Attachment Unit, which was the device that hooked onto the 10base5 cable itself.
Then you run a cable from that device into your network card.
which made routing this big thick cable much easier, and also made it more resilient.
You could stick the cable and Medium Attachment Unit under a floor tile or behind a wall, then then just run the AUI cable to your computer.
a lot of pictures are from here, which is a neat project to build a 10base5 network in the modern day (2012, at least):…
Ethernet was initially developed at Xeroc PARC in the mid 1970s, by several people including Robert Metcalfe.
in 1979 he left Xerox, and formed 3Com, and then he worked with Xerox, DEC, and Intel to promote Ethernet as an open standard, first published in 1980.
Xerox gave up the trademark on "Ethernet" as part of that process (and thank god for that, because the other name was calling it the Digital Intel Xerox standard, or DIX), and it got formalized as IEEE 802.3 in 1983.
Ethernet caught on for a couple reasons, but primarily it seems to have been that:
1. you could buy networking equipment for it from multiple manufacturers, instead of being tied into a specific company's networking ecosystem
2. it was able to adapt to new physical layers
#2 is important because Ethernet started as thicknet, then moved to thinnet, then has stayed with some variant of 10base-T for decades.
The link level protocols being identical meant it was possible to bridge these physical networks together, making it easy to incrementally upgrade, instead of having to throw out your whole network and start over
10base-T, btw, started as an experiment at AT&T to design a network as a star topology. It was called "StarLAN", appropriately enough.
Different types of networks have to be laid out in different ways.
10base5/thicknet and 10base2/thinnet both use a bus layout, where there's one primary path that goes through every computer on the network.
10base-t uses a star, where you have a hub/switch that has a 1-to-1 connection to each computer, individually.
This is more reliable because you can't accidentally take down the whole network by messing up one computer's connection, but it does require more cable in most cases
plus it now requires a hub or switch, which bus topologies never did.
but bus being cheaper meant it was the choice for a long while, with star-topologies only catching on later once prices had dropped.

And the real benefit of a star is that it's better for adapting into a tree.
(that's a weird sentence out of context)
see the nice thing about having your network designed around a star shape, is that you can have multiple stars and then have your switches be smart and route data between them.
you get multiple stars, in other words, without having to wire them all together in one giant bus
which makes it the perfect network design to use as networks grow.
like, say you have a big HQ that's several floors of a building.
with a bus, you have to wire the cable through all the computers on the first floor, then go up to the second floor, then up to the third floor
and then someone knocks over a computer on the second floor and both the first and third floor networks are cut off from each other. so it's very fragile, and tricky to wire.
but with a tree topology, or a star-of-stars, you have a star topology for the first floor, then one for the second floor, then one for the third floor, and all of those talk to one big switch somewhere.
so not only is taking out one computer on the second floor not a problem (it only affects them, because it's a star)
but even if you take out the hub/switch for the whole second floor, it just affects the second floor, not the first or third.
oh, correction! I wasn't correctly thinking about how this works:
if you were on a bus network and lost a computer on the 2nd floor, the 1st and 3rd floors aren't simply cut off from each other, they're cut off from themselves.
remember the terminators?
yeah, the 10base5 and 10base2 network types need terminators at each end of the bus to prevent signal reflection, and when you accidentally break the cable in the middle, there's no terminator there, so the network is almost certainly useless until fixed.
the other big networking technology elephant in the room has already been hinted at: the ring topology.
Why would you design your network as a ring?
and the answer is Token Ring, an IBM standard and competitor to Ethernet. In this case, the network is designed somewhat like a bus design, except that it loops around at the end.
and the reason it does this is because it uses a token-passing method for channel access.

So, see, any network design (other than maybe a fully switched star network where all connections are 1-1) has a problem: talking over each other.
the network connection is common to all the computers, and basically only one device can talk at once, or the signals get mixed up together.
So for ethernet this is handled with something with the complex name of "Carrier-sense multiple access with collision detection" or CSMA/CD, but it's really quite simple. It's very similar to what you do when talking to someone in person, or on the phone.
it's basically:
1. listen to see if someone else is talking. If so, wait for them to stop.
2. now that no one is talking, start talking...
3. but listen to see if someone else started talking at the same time
4. if they did, stop talking, and wait a bit before you start again
importantly the "wait a bit" is randomized and increasing, because if both device always waited exactly the same amount of time, you'd get stuck in that loop where you both start talking, stop, then start again at the same time.
so this is a clever algorithm that lets you use a single common connection for a bunch of computers, but it does have the limitation that your speeds can scale very unevenly with how many computers you have, and how much they're talking
like if you have 50 computers and they all need to talk a little at random intervals, you could easily have them spending most of them time waiting to talk, as they keep stepping on each other's toes.
and you don't have any bandwidth guarantees, either.
You might have a few computers that need to send some small data rarely, but can't because one computer busy broadcasting something huge constantly.
so, with that being how Ethernet handles channel access, how's token ring different? well, it works less like an informal conversation, and more like a group therapy session: Only one device is allowed to speak at a time, and that privilege is passed around the ring
the way it works is that there's a sort of empty-packet that is being continually sent around the network (when it's not actively in use)
Each device gets the packet, and if it has nothing to transmit, simply passes it along to the next computer, so it will loop forever.
but if the device needs to transmit, it will wait until the token is passed to it, then it'll broadcast the message instead of the empty token.
That message will then be passed along the loop.
until the message comes back around to the original device (with value change to indicate it was received), and then the original device starts passing the empty-packet token along again.
So while this is more complicated than the ethernet method, it's also more deterministic.
Ethernet's carrier-sense method involves random backoffs and can easily give you unpredictable speed differences as you change the number of computers and how chatty they are.
token ring on the other hand, you can easily predict how fast a token goes around a network and how adding more computers will slow it down and how long transmitting given data will take. There's no randomness there.
A neat thing token ring can do that ethernet can't is that you can define priorities for different devices.
The token could support allowing certain devices to have preferential access to the network, compared to ethernet's free-for-all.
so you could have networks set up where the server computer can always transmit without worrying that some workstation in the corner will monopolize the network trying to upload GIFs to usenet
physically, token ring networks were set up as star networks.
You'd have multistation access units like this, with connectors for each device, and IN and OUT connections on either end to ring it into other multistation access units.
The network devices hooked in using these neat hermaphroditic (has that name been changed yet? it seems like it might be sometime...) connectors.
These connectors were too bulky to fit onto an ISA/MCA expansion card, so the cards instead would use a DB-9 connector (on the left)
Then you'd use a DB-9-to-token-ring cable (technically called the IBM Data Connector) to hook it into the network.
And since the connectors were hermaphroditic, you could easily connect that straight to the MAU or into a cable from the MAU
Wikipedia points out these were colloquially called "Boy George" connectors.... because obviously the first thing you do with a hermaphroditic connector is start jokingly calling them after the androgynous vocalist of the new-wave band Culture Club.
anyway, networks, won't you?
I'm writing this on a gigabit ethernet network, which involves some fiber links and power-over-ethernet which is a whole other fun tangential subject.
personally, I don't think you can beat the best, first network... sneakernet.
anyway if you enjoyed my long thread about networks, feel free to send me a dollar or two on ko-fi so I can finally fulfill my dream of playing Doom over a token ring LAN
or support me in general with a monthly donation on my patreon:
I really have to stop now because I have to go take medicines and such but I want to point out one final thing:
Token ring networks were designed with the knowledge that breaking the ring would break the whole network, right? so they planned for that.
so with your MAU connections (possibly only specifically the later Controlled Access Units), you actually had two sets of rings, basically a clockwise and counterclockwise one
which meant that if at any point the ring broke, they could simply bypass the fault and go the other way around the ring.
another fun fact:
Token ring isn't just a ancient-past-1980s thing.
the latest token ring standard is a GIGABIT version from 2001.
okay admittedly no gigabit token ring hardware ever surfaced because ethernet had so thoroughly eaten its lunch by then, but IBM and the other companies in the token ring working group were still updating it as late as this century.
Also definitely feel free to correct me on anything I got wrong here. I've read a bunch about networking but pretty much all my practical experience is with 10base-t.
I'm not exactly from the generation where it was the only remaining standard, but I am from the "too poor to afford any networking until like 2001" generation

• • •

Missing some Tweet in this thread? You can try to force a refresh

Keep Current with foone

foone Profile picture

Stay in touch and get notified when new unrolls are available from this author!

Read all threads

This Thread may be Removed Anytime!


Twitter may remove this content at anytime! Save it as PDF for later use!

Try unrolling a thread yourself!

how to unroll video
  1. Follow @ThreadReaderApp to mention us!

  2. From a Twitter thread mention us with a keyword "unroll"
@threadreaderapp unroll

Practice here first or read more on our help page!

More from @Foone

May 15
Given how much right wing opposition there seems to be to forgiving student debt, maybe it's time for a biblical approach to student loan debt?

Every seven years, forgive all debt. Problem solved.
You're also supposed to release the slaves and prisoners so you get prison absolution started as a bonus. Image
I'm just saying, if you keep talking about how this is supposedly a "christian nation" then maybe you should pay more attention to some of the ideas in the Bible, especially when they seem to be more progressive than what you're doing two and a half millenia later.
Read 29 tweets
May 14
So I got a new phone that came in the box for a Nokia 9210.
I say "in the box for a Nokia 9210" because this is not a Nokia 9210.
That's slightly weird.
This is the phone I got.
Read 63 tweets
May 14
I love how the I'm On Observation Duty games mimic the look of both old-school pre-rendered games like Myst and semi-modern throwbacks like Five Nights at Freddy's...

But they're actually just rendered in real time. They're rasterized polygons, not raytraced images.
Technology has moved so far that we used to have to pre-raytrace backgrounds in order to get much higher complexity and quality, and now we just mimic that look because it's easier to render it in real time.
And the game mimics that look by... Basically just never moving the camera.
Read 29 tweets
May 14
so it has one minor flaw which makes it not functional at the moment.
It got uninstalled from its former home with a pair of scissors
flat head screws, of course.
Read 15 tweets
May 13
I hate computers sometimes
so I have a job that has to run some commands using some python programs
and those python programs need libraries, right?
but it could run on various machines! how do we make sure those machines have those python libraries installed?
well it uses a virtualenv & pip, of course!
it creates a virtualenv, then uses pip to install the modules it needs. great.
Read 15 tweets
May 13
fun fact:
"dependence" is a word, so "dependences" is too, and it's a completely different word from "dependencies" which is probably what you actually meant.
guess what typo I just discovered in the naming of a continuous integration pipeline?
oh fun it turns out this was my fault

Read 4 tweets

Did Thread Reader help you today?

Support us! We are indie developers!

This site is made by just two indie developers on a laptop doing marketing, support and development! Read more about the story.

Become a Premium Member ($3/month or $30/year) and get exclusive features!

Become Premium

Don't want to be a Premium member but still want to support us?

Make a small donation by buying us coffee ($5) or help with server cost ($10)

Donate via Paypal

Or Donate anonymously using crypto!


0xfe58350B80634f60Fa6Dc149a72b4DFbc17D341E copy


3ATGMxNzCUFzxpMCHL5sWSt4DVtS8UqXpi copy

Thank you for your support!

Follow Us on Twitter!