OSI: The Internet That Wasn’t
December 7, 2014 6:12 AM   Subscribe

What happened to the “beautiful dream”? While the Internet’s triumphant story has been well documented by its designers and the historians they have worked with, OSI has been forgotten by all but a handful of veterans of the Internet-OSI standards wars. To understand why, we need to dive into the early history of computer networking, a time when the vexing problems of digital convergence and global interconnection were very much on the minds of computer scientists, telecom engineers, policymakers, and industry executives. And to appreciate that history, you’ll have to set aside for a few minutes what you already know about the Internet. Try to imagine, if you can, that the Internet never existed.
posted by jenkinsEar (59 comments total) 28 users marked this as a favorite
 
I came up at that special time where the OSI protocols was something that *was on the final exam*. The protocols aren't used, but the 7-layer model still maps pretty well to reality.

Section 3 "Layering Considered Harmful" of the RFC explains why today it's really just a model though.

tl;dr: IMNSHO When the rubber hits the road, IETF > ISO
posted by mikelieman at 6:27 AM on December 7, 2014 [6 favorites]


OSI will live on forever in the OSI Network Model. All intro discussions of IP networking are legally required to include a diagram of the OSI network model in spite of the fact that IP isn't based on OSI standards. It's some kind of vestigal remnant of protocol wars.
posted by rmd1023 at 6:28 AM on December 7, 2014 [10 favorites]


I remember having to study the OSI model in network class for two sessions after which the professor dropped, "oh yeah and none of that model was actually implemented, here's how networks actually work".
posted by octothorpe at 6:44 AM on December 7, 2014


For my CS undergrad networking course we used Tanenbaum's Computer Networking text, which extensively covered the OSI model. But just a few years later my graduate networking course used his later revision which had switched its emphasis to TCP/IP.
posted by tommasz at 6:51 AM on December 7, 2014 [1 favorite]


What are you all talking about, I use OSI all the time in tandem with my data dictionaries, especially when I'm timesharing.
posted by RobotVoodooPower at 7:01 AM on December 7, 2014 [5 favorites]


One of my first Hard Jobs in coding (well, for my then-current skill level) was writing first a Netbios, then an IP stack in x86 (actually, 8086) under DOS. Being a dutiful and studious young coder, and because the company I was working for was very small and contained nobody who'd written low-level network code either, I went and read up on the seven-layer OSI model and set out to design my code around it.

Some time later, and after really quite a lot of pain - this being in the days before online resources, and never having had any formal training in CS - I managed to disentangle what was going wrong because I didn't know what I was doing, and what was going wrong because the seven-layer model mapped fitfully at best to what was actually needed (TCP/IP wasn't too bad, but Netbios... let us not talk of Netbios).

At the same time as I was causing pain to myself and others this way, I was also going to industry events both to learn stuff and to write up things with my freelance journalist hat on. I remember some very entertaining briefings with Very Clever People from IBM and other Very Clever People from telcos, where they explained patiently why TCP/IP was basically useless and wrong and would never amount to much, while the ISO/OSI stuff was really the bees knees and clearly the shining future of everything online. And Ethernet was all very well for toy computers, but nobody serious would tolerate all that indeterminency and variable latency.

They had to be right, but our little company couldn't begin to do any of that, and what a few people had built seemed to work quite well and our clients seemed to like it, so we'd have to go on doing it wrong...
posted by Devonian at 7:10 AM on December 7, 2014 [13 favorites]


OSI was used a decade and change ago as the backhaul network for switches and between carriers (over SONET, etc). The addressing scheme was entertaining.
posted by graftole at 7:17 AM on December 7, 2014


let us not talk of Netbios

Yikes, talk about the mists of time. All I can remember about NetBIOS is a vague tingling at the edge of my awareness about some transport for IPX and getting networked Doom chugging.
posted by phooky at 7:22 AM on December 7, 2014 [6 favorites]


tommasz: "For my CS undergrad networking course we used Tanenbaum's Computer Networking text, which extensively covered the OSI model. But just a few years later my graduate networking course used his later revision which had switched its emphasis to TCP/IP."

When I went overseas for university, I picked up my textbooks before I left since they were so much cheaper at home. I turned up with the year-old OSI version of Tanenbaum, while the course was all based on the newly-updated TCP/IP edition. I couldn't even give the damn thing away.

Nice article, but the comments are even more fascinating and full of first-hand anecdotes. I guess we know what all the old farts of networking are reading.
posted by vanar sena at 7:22 AM on December 7, 2014 [1 favorite]


The beauty of the OSI model is that it was generalized enough to apply not just to TCP/IP networks, but other types of networking solutions as well (think Storage Area Network protocols, for example). The main problem with the OSI model was that the divisions between the layers often times appeared to be arbitrary and poorly defined. ("Wait, is this protocol I'm looking an implementation of layer 4 or layer 5?" Well, that depends on how hard you squint at it).

Good reading on the history of the proto-internet. It's kind of fascination to look back and think that the stuff we take for granted today was not as inevitable as it seems like it was.
posted by surazal at 7:24 AM on December 7, 2014 [2 favorites]


As background in case anyone is curious what OSI is about: Once it became clear that packet-switching was the future, Ma Bell and (European?) governments decided they would design the protocols in co-operation with equipment vendors. OSI was a top-down corporation-funded design by committee. TCP/IP was a bottom-up pragmatic design based on merit, experimentation and what works - some of them undergrad student hackers. It turned out the committes that designed OSI were so bloated and inefficient, the corporations were so invested in their special interests and proprietary systems, the design specs so complex and impossible to understand.. TCP/IP passed it by before OSI could ever get out of the gate. There is a term for it, I forget, but whoever had the biggest installed base would snowball because everyone would want to connect to the largest network. It was a race to get the protocol stacks installed on as many machines as possible. Vint Cerf was instrumental in this regard.
posted by stbalbach at 7:26 AM on December 7, 2014 [10 favorites]


Great article; great post.

I'm a web developer by trade, but my understanding of the finer details of networking has always been lacking. I, too, encountered the seven-layer model in my larval days, in random books and whatnot. I could never quite make sense of how that model jived with what (little) I knew of TCP/IP networks – the concepts just didn't seem to fit. Glad to learn that it wasn't entirely my fault :)
posted by escape from the potato planet at 7:49 AM on December 7, 2014


There is a term for it, I forget...

It's "network effects". It's why you can't leave FaceBook :)
posted by butterstick at 7:55 AM on December 7, 2014 [2 favorites]


Some of the comments on that article are amazing, make sure to read those as well. (just ignore the inevitable libertarian derail)
posted by any portmanteau in a storm at 8:01 AM on December 7, 2014 [1 favorite]


If you're interested in reading about the future, 9 out of 10 times Spectrum is a better bet than Wired.
posted by GuyZero at 8:05 AM on December 7, 2014 [1 favorite]


Although real session management and not cookies... well, maybe that might have been a good idea.
posted by GuyZero at 8:06 AM on December 7, 2014 [1 favorite]


This article seems to break two cardinal rules - the first is "if you know a subject, don't read journalism about it" and the second is "Never read the comments".

I still can't think of myself as a veteran of the OSI-TCP/IP wars, but I was there and I did get to see some action, so I guess I qualify. It's frightening to see it congeal into far-away history.
posted by Devonian at 8:09 AM on December 7, 2014 [1 favorite]


Great article. I can stop telling people that it was all Al Gore.
posted by Artful Codger at 8:12 AM on December 7, 2014


All I can remember about NetBIOS is a vague tingling at the edge of my awareness about some transport for IPX and getting networked Doom chugging.

I've seen things you people wouldn't believe. 10BASE2 cables strung across the hallway floor. Packet drivers loaded on the edge of EMS memory.
posted by RobotVoodooPower at 8:15 AM on December 7, 2014 [23 favorites]


OSI was devised by committee, but that fact alone wasn’t enough to doom the project—after all, plenty of successful standards start out that way.

Name two.
posted by Tell Me No Lies at 8:27 AM on December 7, 2014 [3 favorites]


I'm amazed at how formative DOOM was for my generation of network folks. So many memories about serial communications, null modem adapters... it supported both the packet-switched and circuit based models for playing, so you had to understand the difference.

Only have 1 COM port? Guess you're playing one on one at the LAN party, instead of 4 player.
posted by butterstick at 8:32 AM on December 7, 2014


stbalbach: the committes that designed OSI were so bloated and inefficient, the corporations were so invested in their special interests and proprietary systems

Sooo... Like the IETF is becoming?
posted by rmd1023 at 8:32 AM on December 7, 2014 [1 favorite]


I was going to make a snarky comment about ATM, but then decided to do some research first and discovered that it's not-at-all obvious whether or not ATM is relevant now.
posted by Slothrup at 8:35 AM on December 7, 2014


Tell Me No Lies: "Name two."

Do USB and H.264 qualify?
posted by vanar sena at 8:38 AM on December 7, 2014 [4 favorites]


As others have noted the OSI Seven Layer Model is a very good teaching tool, although less important now that TCP/IP has crushed most of the other network protocols out of existence.

I still get a lot of use out of it in my "How Does The Internet Work" spiel.
posted by Tell Me No Lies at 8:38 AM on December 7, 2014


Sooo... Like the IETF is becoming?

Yeah that's how the OP article ends, saying: "And OSI eventually collapsed because it could not reconcile the divergent desires of all the interested parties. What then does this mean for the continued viability of the open Internet?"
posted by stbalbach at 8:38 AM on December 7, 2014 [1 favorite]


It's OK, the OSI spirit lived on in the Internet in the form of WS-I, where layers 6 and 7 were broken into a zillion complicated parts. WS-BaseNotification, WS-BrokeredNotification, WS-MakeConnection, WS-PolicyAssertions, WS-Federation Active Requestor Profile, WS-AtomicTransaction, WS-Management Catalog. There was something for everyone! Managed to keep the hands of committee weenies off of HTTP for like 10 years. They're only now coming back and grabbing at it.

Let us also raise a glass to DECnet which had not seven, but eight layers, and was a real thing real people used.
posted by Nelson at 8:45 AM on December 7, 2014 [6 favorites]


I'm amazed at how formative DOOM was for my generation of network folks.

At early cisco it was considered a new technology wasn't fully tested until someone had played DOOM over it. Although in retrospect I think the QA guys just wanted to play DOOM on the clock.

DOOM using MacIP over ARAP will forever haunt my dreams.
posted by Tell Me No Lies at 8:47 AM on December 7, 2014 [2 favorites]


I think it's telling that notably successful standards such as IP, LTE, 802.11 and ADSL have very different genesis stories. I particularly like comparing WiFi and GSM, which started from two very different regulatory stances - the FCC saying "do what you like in that band" and the EU deciding that "Not only do you need a big fat licence, but it will specify to the Nth degree what you will and will not do" - and both ending up wiping out the opposition. Yes, there was a European alternative to WiFi, it was called HiperLAN, it was created by ETSI, and in comparison the Dodo soared like an eagle aloft.

In conclusion: standards are a land of contrast.
posted by Devonian at 8:51 AM on December 7, 2014 [4 favorites]


Do USB and H.264 qualify?

I can't speak to H.264 (and the web page goes from "put out a request for proposals" to a working group with no sense of how complete the proposal was) but USB originally came out of a small group at Intel.
posted by Tell Me No Lies at 9:04 AM on December 7, 2014 [1 favorite]


I grew up on TCP and UNIX and so I guess I bonded to it like a baby duckling as it shaped and solidified my brain's concept of how networking is supposed to work. One day, for obscure reasons you don't care about, I had to install a Sun workstation into a hard core IBM shop. Literally, giant mainframes in the basement, IBM terminals, IBM networking hardware, everything was big blue. Their IT guys were first amused and then confused by my SPARCstation. It was a little computer but it didn't run Windows or OS/2. HOW WAS THAT EVEN POSSIBLE!? They had never even heard of Sun. A few of them had heard of UNIX, but had no idea what it was. Meanwhile, I was trying to network this thing in a Token ring network and OMFG. Trying to read IBM networking documentation was like that scene in CONTACT where you've got all this alien text but none of it makes any sense whatsoever. IBM was like a technology company from the Galapagos Islands... they had independently invented everything and had a completely unique vocabulary. You'd read a paragraph of text and none of the words seemed to mean anything. I eventually gave up and left it to them to figure out.

TCP/IP (and /UDP, I guess) may have issues but everybody using a common language is like 98% of the battle right there.
posted by LastOfHisKind at 9:07 AM on December 7, 2014 [12 favorites]


Dunno anything about networking; is this basically the "how great Amiga/BeBox/MiniDisc/Betamax was, until it got crushed by an inferior yet inexplicably more popular standard" thing?
posted by Halloween Jack at 9:35 AM on December 7, 2014 [1 favorite]


Managed to keep the hands of committee weenies off of HTTP for like 10 years. They're only now coming back and grabbing at it.

My mate Don has been known to observe that those who fail to understand networking protocols are doomed to reinvent them, poorly, over port 80.
posted by flabdablet at 9:47 AM on December 7, 2014 [27 favorites]


Slothrup: I think ATM had a brief window of great usefulness, but then ethernet caught up again.

ATM is a great example of bad compromise - the telco folks want a very small cell size and the IP folks wanted a slightly larger cell size and instead they literally split the difference. So it's sub-optimal for everyone equally!
posted by rmd1023 at 10:21 AM on December 7, 2014 [1 favorite]


I remember conspiring to connect the first privately owned dorm to the UIUC student network using Karlnet wireless bridges, and many many existing network operators at DCL expressing great concern about the difficulty of integrating this mutant network into the existing campus wide AppleTalk and Novell networks.

I just blinked, and said IP only would be fine for now, not mentioning everyone just wanted to use Mosaic to "surf the web", and get their mail with Eudora, on their Windows for Workgroups 3.11 machines, complete with TCP/IP stack from the Euro MS research group at EMWAC.

Two years later everybody at DCL didn't care about those other networks, and claimed they never did care about them. Defeat is an orphan.

By then I had been an ISP for 16 months in a rural county where my sister lived, because farmers use data, and they wanted weather and commodity info cheaper than paying for a FarmDayta satellite feed.

The internet was built as soon as it could be built, with what was at hand and ready to even sort of work. A group working to make the correct solution has to overcome the existing operating standards, because those are constantly evolving in live fire space. Or to paraphrase, IP can span the world while OSI was still tying it's bootlaces.
posted by dglynn at 10:35 AM on December 7, 2014 [4 favorites]


Name two.

Every single one of the structural engineering codes you rely on to keep yourself alive and not-crushed by a few thousand tons of building.
posted by aramaic at 11:06 AM on December 7, 2014 [6 favorites]


the professor dropped, "oh yeah and none of that model was actually implemented,.."
IBM's SNA doesn't count? Because it seemed to have been running on networks for a couple of decades. (Do they still use it? I've been out of that stuff for a while)
posted by MtDewd at 11:23 AM on December 7, 2014 [1 favorite]


Given all the above it's both funny and kind of inevitable that one of the most common connection-oriented applications, VOIP, is more often than not entirely reliant upon using UDP hole punching to overcome the limitations and incidental security provided by NAT. In other words, a pretty grotesque hack that nobody likes is used a billion times a day to subvert another grotesque ubiquitous hack that nobody likes.
posted by George_Spiggott at 11:42 AM on December 7, 2014 [6 favorites]


In other words, a pretty grotesque hack that nobody likes is used a billion times a day to subvert another grotesque ubiquitous hack that nobody likes.

What's even funnier that IPv6, which "fixes" this problem, has been out for 15 years, yet we're still at like 5% adoption.
posted by RobotVoodooPower at 12:02 PM on December 7, 2014 [1 favorite]


There is a term for it, I forget...

It's "network effects". It's why you can't leave FaceBook :)


People often also say, "Worse is Better".
posted by dj glib c at 12:10 PM on December 7, 2014 [1 favorite]


Name two...

Every single one of the structural engineering codes you rely on to keep yourself alive and not-crushed by a few thousand tons of building.

Erm no. Building codes are mostly made up of codifying and attaching quantities to things that are already known. Good carpenters have known for millennia how much load can be safely carried by a 4" x 4" post. Putting this knowledge into a building code means that complete idiots can know this too.

The situation being discussed , the two you're being asked to name, is whether something entirely new, like a new standard for open networking, can successfully done from scratch by large committees, and the answer is... not often. Most often, competing solutions come from small independent teams, and the 'standard' is an after-the-fact recognition of who won.

Remember, a camel is a horse that was designed by committee.
posted by Artful Codger at 12:23 PM on December 7, 2014


As background in case anyone is curious what OSI is about

Or they could just read the article.
posted by Shmuel510 at 12:43 PM on December 7, 2014 [2 favorites]


Oh yeah IBM SNA is totally still a going concern, although over time IBM mainframes have started to acknowledge the Internet exists. But a lot of mainframe applications are effectively standalone data processing setups, so in some ways it's a virtue that they are largely disconnected from publicly routed TCP/IP.
posted by Nelson at 12:52 PM on December 7, 2014


What's even funnier that IPv6, which "fixes" this problem, has been out for 15 years, yet we're still at like 5% adoption.

Not to mention, while we're talking about ATM, it also specifies a priority field on packets that largely covers the sort of low-latency transmissions that ATM/virtual circuits were designed to achieve, if I understand it right.
posted by invitapriore at 1:15 PM on December 7, 2014 [1 favorite]


ATM is very much still alive, though I only know a few places still adding new links. It mainly exists in big multi-nationaL companies/government agencies where ripping out a network to replace it with native ethernet is not worth the trouble.
More amusing is there is at least one IT manager who still defends ATM for its technical purity as the reason he won't transition to ethernet.
posted by bystander at 3:58 PM on December 7, 2014


What's even funnier that IPv6, which "fixes" this problem, has been out for 15 years, yet we're still at like 5% adoption.

Pertinent to this article, when they were coming up with an IPv4 replacement in the 1990s, there were proposals like TUBA (TCP and UDP with Bigger Addresses) that had many advantages, and, in fact, were already being deployed in routers. But it had elements of OSI, and like the article mentions, there was a knee-jerk rejection of anything having anything to do with OSI, so they said no, and pieced together some other suggestions and held their noses and voted for IPv6.
posted by eye of newt at 4:34 PM on December 7, 2014 [1 favorite]


Even if the protocol(s) never caught on, the OSI stack remains an extremely useful framework by which to categorize and contextualize more specific networking knowledge. Simply start with the mnemonic All People Seem To Need Data Processing (or alternately, Please Do Not Take Sam's Poodle Away)...

...and you can expand and use it to explain MAC Addresses vs. IP Addresses, to note the precise differences between Hubs/Repeaters (Layer 1), Switches/Bridges (Layer 2), and Routers (Layer 3), and say exactly why the former are so, well...

You can use it to explain how data is broken up into packets/payloads, marked with addressing information, and sealed into certified (via CRC checksum) media-ready frames, only to be sequenced, stripped, and reassembled at the other end.

You can use the Transport layer to explain the difference between TCP and UDP, and to discuss how the DHCP process must necessarily rely on the latter.

You can use the Session layer to talk about the three-way handshake (SYN/ACK/SYNACK).

I'm in the middle of studying for my certifications at the moment (wish me luck?), and the OSI model has been one of the most fascinating things I have learned about.

On preview: well, crap.

posted by The Confessor at 5:12 PM on December 7, 2014


IPv6, which "fixes" this problem, has been out for 15 years, yet we're still at like 5% adoption.

My main personal objection to IPv6 is that I've quite grown to like not having all my devices get public IP addresses by default.
posted by flabdablet at 5:22 PM on December 7, 2014 [4 favorites]


name two

Last time I checked, the IETF is a committee. So is W3C.
So is the ITU, so GSM and every protocol you use with a "G." in front of it.
And yes, there's a ton of ISO still alive, like many protocols you use that have an "X." in front of it. X.509 if nothing else.
posted by kjs3 at 5:52 PM on December 7, 2014


bystandard: More amusing is there is at least one IT manager who still defends ATM for its technical purity as the reason he won't transition to ethernet.

I remember Token Ring. Tell him good luck with that.
posted by kjs3 at 5:55 PM on December 7, 2014 [2 favorites]


The first mnemonic I learned for the OSI stack (and, yeah, it is a useful learning tool, which is why all intro-to-network lessons are pretty much obligated to include the diagram) was from the bottom up: Please Do Not Take Stupid (alternately Sales) People's Advice. But I'm very fond of the one I discovered recently: People Don't Need Those Stupid Packets Anyways.
posted by rmd1023 at 6:13 PM on December 7, 2014 [3 favorites]


Erm no. Building codes are mostly made up of codifying and attaching quantities to things that are already known.

You are simply wrong. As it happens, I can say this with some assurance, having been on several ANSI-registered committees that feed into the standards which are referenced by the building codes. You'd be surprised at the level of research (nonproprietary and novel, these two concepts can coexist, remarkably enough!) that goes into modern construction.

ASTM, AWS, ACI, NFPA and other meetings are public; feel free to stop by some time and watch us argue about whether we should limit selenium in the galvanizing bath, or if we can get by with just controlling dip speed.

Granted, Internet "engineers" like to talk a lot about moving fast, breaking things, and the virtues of failing often. Fortunately for your daily commute, there are quite a lot of registered engineers who take a more nuanced view.
posted by aramaic at 8:10 PM on December 7, 2014 [8 favorites]


Oh yeah IBM SNA is totally still a going concern

Apparently I am misinformed on this point. My partner is an IBM mainframe guy, and he tells me that the installations he works with have mostly moved to TCP/IP. A particular pain-point for SNA networks is apparently you have to preconfigure all the machines on the network, there's no way a random new host can just join. I guess even in the buttoned-down mainframe world that's too much hassle.
posted by Nelson at 8:33 PM on December 7, 2014


A particular pain-point for SNA networks is apparently you have to preconfigure all the machines on the network, there's no way a random new host can just join. I guess even in the buttoned-down mainframe world that's too much hassle.

Maybe the base VM host (or zOS or whatever they're calling the hypervisor/bare metal layer these days), but you can spin up & down VMs with all sorts of interesting ad hoc configurations.
posted by kjs3 at 9:29 PM on December 7, 2014


Granted, Internet "engineers" like to talk a lot about moving fast, breaking things, and the virtues of failing often.

No, no. [br][o|e]ntrepreneurs might like to talk about those. Network/Internet engineers very rarely do.
posted by Juffo-Wup at 10:23 PM on December 7, 2014


Name two.

i can name MIDI as one
posted by pyramid termite at 2:54 AM on December 8, 2014 [1 favorite]


Last time I checked, the IETF is a committee. So is W3C.
So is the ITU, so GSM and every protocol you use with a "G." in front of it.
And yes, there's a ton of ISO still alive, like many protocols you use that have an "X." in front of it. X.509 if nothing else.


So apart from TCP/IP and HTTP and ssh and UTC and HTML and Frame Relay and mobile phones and security certificates, what have the Romans ever done for us? Nothing!
posted by flabdablet at 4:39 AM on December 8, 2014 [4 favorites]


Yeah, brontrepreneurs were actually quite nimble, despite their public image as lumbering oafs stuck in the Jurassic.
posted by No-sword at 4:42 AM on December 8, 2014 [1 favorite]


rmd1023: All intro discussions of IP networking are legally required to include a diagram of the OSI network model…

[ERROR TYPE: SYNTAX. MESSAGE: OSI-NETWORK-MODEL-AS-SEVEN-LAYER-DIP ANALOGY MISSING.]
posted by wenestvedt at 8:27 AM on December 8, 2014


« Older America's Worst College... Or Is It?   |   Face Control - A Moscow Travelogue Newer »


This thread has been archived and is closed to new comments