Slashdot Mirror
NSFnet — 20 Years of Internet Obscurity and Insight
coondoggie writes "The National Science Foundation (NSF) reissued the words that started the Internet revolution 20 years ago today:
'The NSFnet Backbone has reached a state where we would like to more officially let operational traffic on.'
That was the email sent to users of the NSF's fledgling NSFnet to announce that the network's backbone had been upgraded to a 'blazing T-1 speed.' NSFnet was created by NSF a few years earlier in an attempt to create a computer network similar to the Department of Defense's ARPANET.
When the original six-node, 56 kilobits-per-second NSFnet backbone went into operation in 1986, NSF made the decision to allow any academic, governmental or commercial entity to hook up to this network of networks. Within a few weeks of going online, traffic on the new network began doubling every few weeks. The network's backbone of core 56 kilobits-per-second connections were considered fast, but they were not fast enough to satisfy the demands of all the new users who were coming online, according to the NSF."
You are probably talking about FidoNet.
"Alcohol, Tobacco, Firearms, and Explosives" should be a convenience store, not a government agency.
For those of you who have really never heard of it before, the National Science Foundation Network (NSFNet) was a major part of early 1990s Internet backbone.
Basically, here's what happened: following the deployment of the CSNET, a network that linked academic computer science departments, in 1981, the NSF aimed to create an open network allowing academic researchers access to supercomputers. In 1985, the NSF began funding the creation of five new supercomputer centers: the John von Neumann Center at Princeton University, the San Diego Supercomputer Center on the campus of the University of California at San Diego, the National Center for Supercomputing Applications at the University of Illinois at Urbana-Champaign, the Cornell Theory Center at Cornell University and the Pittsburgh Supercomputing Center. The NSFNet connected these five centers and allowed access to their supercomputers over the network at no cost. The NSFNet went online in 1986, using a TCP/IP-based protocol that was compatible with ARPANET, as a backbone to which regional and academic networks would connect. It experienced exponential growth in its network traffic. The original 56- kbit/s links were upgraded to 1.5 Mbit/s in 1988 and again to 45 Mbit/s in 1991.
When did privatization begin? Well, In the early 1990s, commercial organizations connecting to the Internet had to sign a usage agreement directly with NSFNet to gain access to large parts of the public internet, regardless of what Internet Service Provider they purchased Internet access from.The original 56-kb/s backbone was operated by the supercomputer centers themselves with the lead taken by Ed Krol at the University of Illinois at Urbana-Champaign. PDP-11/73 Fuzzball routers were configured by the University of Michigan and statistics collected by Cornell University. From 1987 to 1995 the NSFNET was operated on behalf of the NSF by Merit Network, Inc., a non-profit corporation governed by public Universities. On April 30, 1995, the NSFNET Backbone Service was successfully transitioned to a new architecture, where traffic is exchanged at interconnection points called Network access points.
However, some aspects of NSFnet have been controversial. For much of the period from 1987 to 1995 there was concern by some Internet stakeholders, following NSFNET's opening up the Internet, over the effects of privatization and the manner in which IBM and MCI were given a perceived competitive advantage in "leveraging" federal research money to gain ground in fields that other companies were allegedly more competitive in. The Cook Report on the Internet, which still exists, evolved as one of its largest critics. Other writers, such as Chetly Zarko, a University of Michigan alumnus and freelance investigative writer, offered their own critiques.
I hope you've enjoyed reading this history of NSFnet as much as I enjoyed researching it! (Using the Internet!)
Suggested moderation: +1 Informative, +1 Insightful.
Fidonet and all the echomail networks which appeared after that, using the very same protocol.
Fidonet used zones from 1-6 (1-North America, 2-Europe, 3-Asia(?), 4-Latin America.. etc), each BBS had an unique address, such as 4:804/3 etc.
Fidonet addressing was organized as ZONE:REGION/NODE or (less common) ZONE:REGION/NODE.USER.
Other networks used unallocated zones, such as 39-Amiga Net, 20-Lusonet, 65-Mufonet.. etc.
In Brazil there were a number of nationwide Portuguese-speaking networks too: 12-RBT, 30-Syncnet, 100-Canal 100, 120-AmigaNET-BR etc etc.
I remember there was even a e-mail-like service (called netmail), so you could send a private message to JoeUser@12:345/6.
There are so many histories.. Such as the power struggle between Brazil vs Mexico (in ~1993) for being the main Latin America hub of Fidonet. It was quite a dirty war, at some point Mexico stopped routing messages from Brazil and things like that. -- I remember in the end Mexico "won" but both brazilians users and sysops were so pissed off that everyone migrated to RBT and Fidonet in Brazil suddenly died (later it recovered, but RBT remained the most active network in this country).
That was the start of the IP Internet. The 'Matrix' (as John Quarterman called it then) has already been coming together for a decade as a loosely coupled set of independent networks using different protocols internally. I remember using DecNet in 1980 to chat and e-mail between Ireland and California. I was also on several interconnected BBS (think FidoNet, and later CompuServe) and AMPRNet in '82 (only 1200 baud but we built our own infrastructure and did not need an ISP :-)
IP made things much easier when we started using the same protocol on all the nets.
Actually, what summary says is that it started out as a 56k backbone and then when they noticed the demand they upgraded to T1.
A bunch of folks are bemoaning the 56k number, as it seems rather an odd rate. Awright younguns, hop up here on Uncle MigraineMan's lap while he tells you a story ...
... and was a circuit board about the size of an ATX motherboard. Not wanting to transmit all those pesky bits, another bunch of smart lads realized that the human ear isn't a linear device, so they encoded the 14-bit linear samples using the dreaded u-Law encoding table. That made each sample a more manageable 8-bit value.
... anybody?
Back when communicating between two distant places involved two tin cans and some wet string, some mighty smart folks invented digital telephony. First, they decided to sample the voice audio at 8kHz - after all, they were only obligated to deliver audio bandwidth in the 300-3000 Hz range (affectionately referred to as "three hundred to three K C" back in the day.) You might be surprised to find that a 14-bit analog-to-digital converter that sampled at 8kHz was quit the engineering marvel
Whelp1: But Uncle MigraineMan, what's that got to do with 56k?
Now just settle down a bit. [MM sips from pocket flask.] So the bright young engineers decide that 24 is a nice round number, so they grouped 24 voice channels together into a Digital Signal 1, or DS1. If you follow with the math, you multiply 8-bits by 8000 samples per second to get 64000 bits per second, then by the 24 channels to get
Whelp2: 1,536,000. But Uncle MigraineMan, everybody knows that a DS1 is 1.544Mbps!
That's right. One of them bright young engineers realized that they couldn't tell head from tail with all the voice channels looking the same, so they added some bits to mark the start and end of the DS1. That brings us up to the current line rate.
After a while, the phone company - and note that I said "the" phone company, as there was only one at the time - started using these fancy DS1 signals as connections within their network. They started noticing that when a bunch of calls on a DS1 were silent, sometimes the DS1 equipment would drop out, causing many disgruntled customers. And as I always say, if it affects the revenue stream, it gets immediate attention.
The bright young engineers studied the problem, and discovered that a long period of silence could cause a long string of all-zeroes in the fancy DS1 signal, causing the terminal hardware to think the line had been cut. To remedy this situation, the bright young engineers decided to add some "1" bits to the audio channels to maintain what they call "ones density." That's a fancy way of sayin' they limit the number of consecutive zeroes so the fancy DS1 line equipment doesn't get confused. They decided that, since this is voice audio, and they've already compressed it with the dreaded u-Law code, no one would notice if they "stole" that least-significant-bit and made it always a "1". It is, after all, "least significant." Who's going to miss it?
Whelp1: So there's only 7-bits of usable data in each voice channel? That's nuts!
Well, it made sense at the time. Eventually, computer usage forced the phone company to upgrade it's equipment to support "clear channel" transport, instead of the "robbed bit" format. That caused a whole passle of problems during the transition. Ultimately, something called B8ZS was pretty much universally adopted. Another day, I'll tell you a scary story about something called ZBTSI. Now y'all run along.