It's time to mark six decades of computer networking Next week I am heading to Edinburgh University, where I did my PhD back in the 1980s, to give a lecture as part of the events celebrating 60 years of Computer Science and Artificial Intelligence at Edinburgh. Having decided that I would speak about networking (stick to what you know, after all), I realized that there was a lot …
One of my favourite quotes was by one of the guys that came up with TCP/IP, he said "I was there when it was invented, I know down the electrons in the wires what is going on but when I look at the amazing networks we have in the world it all seems so magical how it all just works.".
Humble and awed by what you invented and gave to the world, that's the mark of a true techie, part scientist and part wizard!
I was gobsmacked when I managed to get two computers to talk to each other down a null modem serial cable when network cards were > £300. Before that networking seemed to expensive to be part of personal computing. After that open standards seemed like the way to nirvana. And they still are.
My first job (1968) was with a university department that was funded by ARPA - "The Socialization and Rapid Acculturation of Native Cultures" - all loaded by punch cards into the mainframe. Definitely not networked.
A bit later, I worked in Northern Virginia had an IMP (Interface Message Processor) as part of a testbed in the mid 70s. It was loaded with paper-tape and had the normal plethora of toggle switches to enter codes into memory.
A bit later I ended up with Berkeley and watched the incredible tsunami of innovation and interconnections that these networks opened up. Still amazed at the technologies and dismayed at the commercialization.
This post has been deleted by its author
I'd add "Computer Networks and their Protocols" by Donald Davies, Derek Barber, W.L. Price and C.M. Solomonides, Wiley, 1979, for an even earlier tome by one of the inventors of packet switching and his colleagues at NPL.
I was in meetings with Derek Barber in the 1980s, and he was always pretty reluctant to accept that TCP/IP had "won" the protocol wars.
(Lots to enjoy at https://ethw.org/Oral-History:Donald_Davies_%26_Derek_Barber )
And any history of networking in the UK must pay respect to Peter Kirstein, who brought the ARPANET to London in 1973.
Include the Cambridge Ring
The Cambridge Ring had one facility that I seriously miss with IP. If A had a connection to B and B had a connection to C, B could splice the two connections together so A & C communicated directly from then on, without A or C having to do anything or even know about it. Absolutely brilliant for service broking without fuss.
That was course material for me too, just checked on my bookshelves and its not there anymore. I must have gotten rid of it for not being 'relevant' enough. Of course Stevens et al. are still there.
I remember our sysadmins getting really excited about network services like NFS in the early 80's and I was like "yeah, why wouldn't that be a thing?". We take a lot of stuff for granted.
Was never a fan of the OSI model. It seemed to be at odds with everything and everyone. No, no, it should be like THIS dammit!!
The OSI reference model was good navigational and educational aid and hence has relevant today. Personally, it would be good to review and update based on 40+ years of experience, where we have moved beyond telnet, file transfer and email.
The objectives OSI were good and the Internet does embody these.
The OSI services and protocols were well thought through, but at times the “absolutists” and need to accommodate differing vendors viewpoints was irritating and ultimately tainted OSI. interestingly, some of the newer (ie. Post circa 1990) work on the Internet Protocol Suite either originated in OSI or has benefited from tools developed to support OSI standards development.
"messaging between worldwide users on VM "
I was going to say that only IBM could come up with an e-mail solution so weird that it involved redirecting your virtual card punch to someone else's virtual card reader via the spooling subsystem, but it would be unfair to the people who came up with UUCP.
"No muss and no fuss"
Not once you've sorted things out with the breakout box & determined what flow control and parity are being used.
IME there's no sort of computer interconnection where muss and fuss can't be introduced given enough determination and lateral thinking.
I was talking about the kit in and around chez jake, much of which I've built. I was not talking about the industry at large. The only thing I have that isn't standard is an APC UPS that I keep running for contractual reasons. That contract expires shortly, at which point the APC will get recycled with great conviction.
"Standard were RS-232 terminal ports"?
What are you talking about?
It used to be a major part of my job for years to study the pin-outs of the serial ports of various computer systems, and then slave away firstly with line monitors, null-modems, gender-benders, break out boxes, and then with with hot soldering irons and multi core cable to make specific cables, and sometimes even break out the protocol analyzer in order to get RS-232 links working and reliable between different systems.
The problem was that there was a standard, but it was so all-encompassing that vendors selected different subsets to implement, and the standard did not really even go down as far as the physical connectors (and there were some strange ones out there - IBM, I'm looking at you for your Modu 10 pin rectangular connector, and also Acorn for their 5 pin DIN RS-423 connectors).
Quite. In the 1980s we had a good business making a network that linked RS232s together. As one customer said, "we have N devices with serial ports and used to have N-squared problems linking any to any, now we just have N problems linking each one to the network." Later we made nodes with parallel interfaces and file-sharing software.
While the TCP/IP folks went connectionless to avoid having to keep information on flows, we went connection-oriented to avoid having to keep information on all the reachable addresses.
A mass-produced UART and a D-sub connector were a lot more "standard" than a flat cable snaking out between two panels with no connector at all.
But yes, it was a nightmare of competing "standards". I own some test equipment (probably ex-US Army) that has true D-sub 9 connectors for serial communications. That's a D-sized connector with 9 pins in a single row down the middle. The pin-out seems to have been assigned at random. I have absolutely no idea why they built it with such a non-standard part ... In about 1990 I called Amphenol for spares, they told me that they made them for a limited time in the early 1970s for a government contract, and they sent me a box full of old stock, gratis (individually wrapped, complete with pins, hoods & hardware). I probably still have a couple dozen or so of each (male and female) in my junk collection. I've never seen 'em anywhere else.
>What are you talking about?
Didn't say they were all wired the same and thus having a break out box and an ability to wield a soldering iron were useful to a "programmer"(*) :)
I wasn't aware of the IBM special Modu 10 pin connector.
(*) For those fammiliar with the saying "Beware of programmers who carry screwdrivers" and variations there of, a soldering iron isn't a screwdriver...
Current loop was all but gone once UNIX was becoming widely available at Universities.
The DEC VT-50 (1974?) was *optionally* available with current loop (20mA), according to the advertising of the era and the manual that came with them. I have four or five squirreled away, just in case (two are still out on loan to the The Computer History Museum ... I should probably just sign them over to them permanently). Most of the other early DEC terminals were also available with the current loop option, but I don't have a handy list, nor any physical examples. My point being that it was an option, not the standard, as early as 1974.
Not certain why you're calling out a line editor in this conversation ... it goes without saying that the operator had to know whatever editor came with the system at hand.
We had tetetypes with current loop converters on several of the systems I used, probably in the late '70s or maybe early '80s. I have no idea how old they were, or where they had been previously used, but they were connected to RS-232, but not to UNIX systems (Harris, if I remember correctly).
Current loop devices were required for telex systems, and I think that this is where many old hard-copy devices came from.
Bell Labs, being mainly a telecoms company even though they initially developed UNIX almost certainly used them.
The reason why I called out a line editor as there was some overlap between people only having hard-copy terminals on UNIX systems, and VI, as you of all people should know.
For my first year of using UNIX, we used ed even though we had VDUs, mainly because the Newbury VDUs (UK manufactured terminals) were too dumb to do cursor addressing (they really were dumb terminals, or to resurrect an old term, glass-TTYs). Fortunately, our sysadmin knew some people at Queen Mary College in London, and got a copy of em (Editor for Mortals), which could work even with the limited commands available on the terminals. We may have been able to run ex with the visual components removed (I can't remember whether the BSD tape that Durham had [mainly for Ingres] was late enough to include ex, but the one I had when I worked at Newcastle Poly. did include ex), but our PDP-11/34s (at both places) could not run VI, even with the Keele overlay loader.
To be fair, Konsole (and most other KDE stuff) was programmed by kids who never used a Model 33 in anger. Or at all, probably.
Are you sure your tape handler has a punch, and is not just a reader? You should be able to manually put the unit into punch mode, where it will take whatever you send it as input and punch the result ... which might be gibberish, if your hardware settings are incorrect.for the data type, but you'll probably get something.
If it's not a 33, is it five, six or eight hole tape? Manufacturer & model number?
My college job was repairing Model 33s that got used in anger.
I had an endless supply of epoxy and 16AWG solid wire, which I used to repair the plastic covers slammed into by undregraduate fists, as they learned how to write code that didn't compile. The transparent window over the typing area was a favourite target
I recently re-implemented UUDECODE in BBC basic to allow me to transfer ROM images over RS-232/RS423 from one of my Linux systems to an SD card on one of my referbished BEEBs (for some reason, I just could not get a modern(!) KERMIT implementation to talk to the KERMIT software I have for the BEEB).
It's amazing what you end up doing when you have time on your hands and are bored.
Still, it was fun.
What really, really worries me is that I not only understood that but that I could possibly still do it all after a bit of ersfreshing of the wetware cores.
That and more.
Man, the things we had to know!
Is it better now that everyone can run a super-computer with his thumb?
This post has been deleted by its author
"for some reason, I just could not get a modern(!) KERMIT implementation to talk to the KERMIT software I have for the BEEB"
If you can be arsed, ask over at https://www.kermitproject.org/ ... I'd be interested in the answer (if any), might come in handy someday. (There was a bug from a couple years ago where -T and -B (text or binary transfer command line switches) were accidentally reversed, but if I recall correctly that was for E Kermit, the embedded version.)
There is not much BEEB gear on this side of the pond, but I have friends over on your side who still dabble.
Wrong question?
The more important question would be: "Why not?"
If it works for thousands of programs, millions of people and quadrillions of boxes, switches, networks and other thingys, why bother changing it?
There is also the "how the Hell *do* we change it?" issue. See "removing IP4 to implement IP6-only" for details. :)
Tanenbaum seemed to do a good job of catching both the focus of the moment in the various editions and the changing view of the past, making each edition worth keeping as there was little real overlap.
Perhaps Bruce could consolidate his researches into circa 1960 data networking into a “prequel” volume…
SNA is far from forgotten. It is still used daily by pretty much anyone who uses mainframes and/or mainframe technology. That would be your favorite financial institutions, your alma mater, most space programs, your favorite automotive and aircraft companies, major mining and drilling companies, shipping companies, and your government. That is just for a start.
DEC's management quite simply squandered the franchise. They fucked up. There is no pretty way to put it.
Well, yes, but nowadays SNA runs encapsulated over IP. IBM itself inverted their internal networks in the mid '90s so that SNA ran over IP, rather than the rather bizarre situations where they had TCP running point-to-point over the SNA transport layer in some parts of their network, and recommended that their custoimers did the same inversion. It meant that our island of TCP/IP in the UK AIX Support Centre was able, finally, to talk to the development labs in the US and Canada directly for fix delivery, rather than using the mainframe networks to have emergency fixes sent to us. Running FTP (or even HCON/3270) to a local mainframe to download files sent from Level 3 support was quite an experience, one that I would prefer not to have to repeat.
This was back when TCP/IP and Ethernet was still quite foreign to IBM (the network inversion happened when IBM outsourced their global network operations to AT&T, as I understand that AT&T wanted to have as little as possible to do with SNA).
Funny but true: IBM's WorldWideNetwork wasn't overseen with IBM computers ... it was overseen by Sun Microsystems computers which were re-badged by a company called Network Equipment Technologies after their software was installed on them. IBM Field Service knew virtually nothing of UNIX, TCP/IP or even modern email in that era (early to late '80s), and tended to look down their noses at those of us who did. Funny thing is I (in my jeans and T-shirt) was making more per hour than any three of them (in their three-piece Armanis) ... Probably fair, given that I also knew SNA along with the UNIX side of things.
Eh...All the makers of minis and superminis got overtaken by [single and multiple] Pentii. It's not clear that any of the proprietary manufacturers could have done anything to survive the rise of commodity motherboards. Standard, non proprietary hardware (and, later, POSIX) levelled the playing ground, something that DEC and their bretheren couldn't deal with.
"Have they really improved?"
Yes. Today I can easily go years of 100% uptime and connectivity.
Back then, going three days in a row without a crash and/or the link dropping would have been noteworthy.
"Complexity itself now is becoming the enemy."
The hardware and the relevant parts of the OS+network stack haven't really changed all that much complexity-wise in decades.
The hardware and the relevant parts of the OS+network stack haven't really changed all that much complexity-wise in decades
go build one of the many modern fabrics from the likes of arista, juniper (evpn), Cisco (aci)
still pushing packets about but the way we do it today is vastly more complex than it ever was before.
Yes, that bit. Maybe try parsing it after reading it?
Apparently you are unaware that one can still connect to the existing Internet with a stack from the early 1980s, and use that connection[0] to do useful work in today's world. In other words, as I said, "The hardware and the relevant parts of the OS+network stack haven't really changed all that much complexity-wise in decades".
[0] Note that I don't recommend doing this unless you have a good working knowledge of firewalls.
Apparently you are unaware that one can still connect to the existing Internet with a stack from the early 1980s, and use that connection[0] to do useful work in today's world. In other words, as I said, "The hardware and the relevant parts of the OS+network stack haven't really changed all that much complexity-wise in decades".
yes TCP/IP protocol still works as it did many moons ago which means the OS and network stack of the sender and recipient from 30 years and more ago can still connect to systems old and new that are complaint with the protocol.
the network between the sender & receiver is very much different though which was the point of my comment about modern fabrics being very different today. It looks like you didn't understand that nuance.
>Have they really improved?
The reason for the header checksum was not to detect transmission errors (the datalink protocols have their own checksums for that purpose), but to detect memory corruption in what we would now call "routers". The ARPANET crash of 1980 was ultimately determined to be the result of memory corruption in one device (IMP) that resulted in the propagation of erroneous routing information to others.
Of course, we now all have ECC memory, don't we? Don't we...?
It wasn't really a crash, all the hardware remained up and running and could talk locally. It was just remote IMP to IMP traffic that was affected. Basically, some b0rken hardware caused a flood of messages that the network was unprepared to deal with. In essence, it was the first denial of service attack, but the network actually DOSed itself. The gory details were immortalized in RFC 789.
http://www.faqs.org/rfcs/rfc789.html
Note that the 'net was only down for about four hours.
Well, my Win-7 box has been operating 408 days continuously running four instances of Prime95, so I would say "yes, reliability has improved a lot".
It's a 2013 box and has been going for a decade so reliability is *really* better and has been for years.
I vividly remember trying to download a 60MB (yes, *MEGA*-byte) video file for my wife and it failing repeatedly and needing to be restarted for over a week.
Today, I can watch multi-GigaByte movies as streams with every expectation of zero interruptions for buffering.
I can watch them while P95 runs in the background, too.
Today's tech is incredibly impressive.
This post has been deleted by its author
"Globe Al Warming"
Maybe if he didn't fly around all over the globe in Big Jets bitching about the climate he wouldn't be contributing to the problem? Same goes for the rest of the global warming conference attendees. Hypocrites. You know who you are ...
i love reading about computing history & especially networking history.
its a suprise to many that concepts of what we do today originated in the late 1800's tdm having been used from the 1870's as an example.
https://en.wikipedia.org/wiki/Time-division_multiplexing#History
how about fax machines being invented and used in the 1800's
https://en.wikipedia.org/wiki/Fax#History
we take for granted that things just work, i find the story behind it all fascinating.
Every time i look at (not into!) a fibre optic, I cannot but wonder at all those pulses/flashes of light.
Not 1,000,0000 pulses per second, not 1000,000,000 pulses, but 40,000,000,000 or even 100,000,000,000 pulses per second. And every single one one of those pulses counts. It means something to someone. Something completely and utterly different than the previous or subsequent pulse. They have all been assembled in a precise order, and at the far end of the glass strand, all get dis-assembled and added into new streams of hundreds of thousands of millions of pulses per second all precisely ordered and going in multiple different directions. And in an instant, that happens multiple times, and those flashes get propagated around the planet and end up exactly where they were intended to be,
And then the pulses get precisely disassembled from that fibre, reconstructed into a binary sequence, which in turns gets reconstructed into words, or a picture, a song or a video or details of a medical procedure or drug that saves someones life.......
It's just amazing. Gobsmacking. Imagine trying to explain how it works it to your grandmother, or great-great grandmother.
And humanity has achieved this in just 60 years. Unbelievable. Thank you to the visionaries who thought it out and made it possible. Humanity owes you more than you will ever know. And i and fellow commentards owe you our entire careers.
And everyone else takes it for granted, whinges when it doesn't work, and owes you Facebook. Oh well.
This post has been deleted by its author
I would never have tried to explain it to my mother.
Explaining it to my *sister* is like an exercise in talking Tagalog to an Icelander when either of us understands the language and I don't do Icelandic.
On the other hand, my wife took to it exceedingly well.
It isn't all culture-lag or time-locked milieux.
On the exponential growth of these technologies, the last few years have shown us that a couple of obnoxious multi-wealthy types can use their commercial enterprises as
massive brakes on innovation and technological change. Suing another company for having rounded corners on their devices is not exactly conducive to getting us off this
planet and walking among the stars any time soon.
We used to be ever so much better than that.
I worked on 2.4Gb FO chips in the late 80s - one of my work mates had a 9.6Gb FO receiver working from some 'test' chips we had made. I could send 2.4Gb/s 10km with an error rate of 1 in 10**14 when pretty much the biggest hard drive you could get was 40MB. Kinda weird thinking you could fill your hard drive faster than a dial up modem could connect!
And 33 years later we have twitter having to limit the number of reads of small paragraphs of text its users can share. You can explain this to the 'leaders' of modern high tech let alone your grandmother.
I tried to join the Fediverse a couple of times, but each time everything went black. Except the white text that burned my eyeballs. Sorry, but for some of us black backgrounds are just not usable. Also, tasteful thin grey borders are far harder to see on black than on white. There is a reason we print black ink on white paper and whiteboards took over from blackboards. It's not so bad if I can turn the white bits a restful shade of green, but I guess that an individually customisable UI is low on the developers' priorities.
So when the Fediverse is big enough to be egalitarian and offer us lusers a little normality, children like me are sadly excluded.
All the early experiments with networking had two properties -- they could be unreliable and they were expensive. Although I started developing network products in the 1980s they really were not ready for prime time. The first real break happened with StarlAN, native networking on phone cable that didn't use phone technology. This was the ancestor of modern wired networks and the mass market. The breakthrough came when network adapters became cheap and compact enough to fit on a computer motherboard so became standard on all new computers.
Wireless LANs went through a similar development cycle. There were early attempts at wireless LANs for offices in the 1990s (e.g. Xircom'a Netwave) but the technology but it took to the mid 2000s for wireless to become universal, cheap and reliable enough to be fitted as standard to new laptops -- and then to just about everything.
I do remember a relic one of my employers depended on - Bitnet. Because It's Time net. EBCDIC encoding, IBM protocol based store and forward e-mail processing. Our end node was a DEC MicroVax 3500. And all the site-to-site links were 9600 Baud since the e-mails were text only, no binary attachments.
There was a European version of BITNET called EARN, it was still around in the 1990s. I was using it to connect from the UK to CERN at around the time CERN created the WWW. The transatlantic gateway between EARN and BITNET was 9600 baud.
The reason the WWW didn't exist before 1990 wasn't that nobody had the idea (lots of people had the idea) but that it was not feasible or economic to implement the WWW with networking costing what it did back then.
Great topic. Just an FYI, the mention of building the network to survive large scale enememy attacks can be found in Paul Barons Rand Corp papers. The are a great read in general. Here is a link to one https://www.rand.org/pubs/research_memoranda/RM3420.html
Very understandable, they also demonstrate how some of the decisions were made at the time before we had better math to describe and handle networks (most of that came later).
" the mention of building the network to survive large scale enememy attacks"
Not "the network", rather "a network". The paper you cite is an abstract on networking in general, not a design study for The Internet. The fact of the matter is that (D)ARPANET, and later what we now call "The Internet" was never intended to survive such an attack. It just plain wasn't in the design specs (such as they were).
In actual fact, the US Government already had such a high probability of survival communications system. The networks that were designed to survive nuclear attack included the "Minimum Essential Emergency Communications Network", or MEECN, and the prior "Survivable Low Frequency Communications System" or SLFCS, Besides, if you use an ounce of common sense, it only stands to reason ... no military would design a command and control system that inherently wasn't securable, and the Internet was not then, and still isn't securable.
Boiling it down to basics, the (D)ARPANET was just a research network designed to research networking. The "survives nukes" myth came about much later ... The cold, sad reality is that the only reason it was built to be resilient is because the available hardware was really, really flaky.
Another hagiography of the ARPANET/IETF axis without acknowledging the true inventor of internetworking, Louis Pouzin. The original ARPANET was not an internet, just a packet-switched closed network. Internetworking means the ability to send information across multiple independent networks without their cooperation. That is, the internet's packets are the payload of an underlying network who doesn't care about the payload. Pouzin saw the early ARPANET and in 1972, at INRIA in France, developed Cyclades, an experimental true internet. A paper he wrote about it described the key aspect where it was layered on top of telecom networks. Metcalfe and Cerf leaned about this and in 1974 came up with The Transmission Control Protocol (IP split from TCP in 1978, in version 4), a bad interpretation of Pouzin's work. A change in French administrations, in favor of the post office (who opposed it), shut the Cyclades project down so it was largely forgotten.
The author says there is no mention of "the Internet being able to withstand nuclear attack". This may be true for the evolved Internet but there is good reason to imply it from its origins in packet-switching. Once packet switching was available its potential vs the alternative technologies of the time (telephone networks and short-wave radio) was attractive as a more survivable solution under nuclear attack due to distribution and redundancy. This is best described in a review paper "The Beginnings of Packet Switching: Some Underlying Concepts" by Paul Baran in the IEEE Communications Magazine of July 2002, pp 42 - 48. The most relevant paragraphs are entitled "Cold War Background" at the beginning and "Disclaimer" at the end. This also reinforces that the term "packet-switching" was coined by Donald W Davies of the British National Physical Laboratory principally to distinguish it from Message Switching as used in telegraphy and teleprinter working.
"but there is good reason to imply it"
Again, the papers you cite were abstracts on networking in general, not specific instructions for building (D)ARPANET or the later Internet.
Why is it that humans hang on to myths so tightly, even when people who were actually there at the time say "No, it did not happen that way"? But don't take my word for it, instead why not ask The Internet Society? I'm pretty certain that they would know, if anybody does. In fact, here is a document called "A Brief History of the Internet", published in 1997 and signed by the likes of Vint Cerf, Bob Kahn and Jon Postel (and others). See particularly footnote 5, which states:
"5 It was from the RAND study that the false rumor started claiming that the ARPANET was somehow related to building a network resistant to nuclear war. This was never true of the ARPANET, only the unrelated RAND study on secure voice considered nuclear war. However, the later work on Internetting did emphasize robustness and survivability, including the capability to withstand losses of large portions of the underlying networks."
And again, this was because the hardware of the day (including the links!) was really, really flaky. No nukes required. Bits of the 'net fell over all by themselves. Regularly. It didn't help that grad students were hacking on it 24/7 in their attempts at wresting a higher degree out of the kludge. We Broke Shit. Some would say it was our job ... For example, In late 1977 I managed to take down all the PDP10 kit at Stanford and Berkeley with a kernel upgrade. Effectively split the West coast DARPANet in half for a couple hours. Not fun having bigwigs from Moffett and NASA Ames screaming because they couldn't talk to JPL and Lockheed without going through MIT ... At least the network managed to handle the damage and route around it :-)
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
