Reflecting on the 20th Anniversary of NSFnet, Internet Origins

The NYT and news.com have up an article looking back at the NSFnet's influence on the development of the internet. From the National Science Foundation's gamble came the TCP/IP standard we know and love today; when NSFnet was shut down in 1996 it was apparently connecting some 6 million computers. The piece also talks about the (sometimes tense) relationship between private and commercial interests. "The Internet 'was an alien concept to the communication industry when it began growing.' While there was no risk for MCI, which was then an upstart trying to gain ground on AT&T, that was not true of IBM. The company played a crucial role in the development of the Internet, and it did so despite the fact that the new network was a direct competitive threat to its multibillion-dollar communications networking business, based on a competing standard known as Systems Network Architecture, or SNA."

Comments

Why read this article??

[gskouby]

Why read the article when you can get the information first hand? http://www.nsfnet-legacy.org/
Check out the archived videos - this should be required viewing for everybody.

Al Gore meme update required

[QuickFox]

FTA, surprising information that slashdotters need to know:

In the 2000 election, Al Gore [...] was derided by opponents who claimed that he had said he "created" the Internet. But many of the scientists, engineers and technology executives who gathered here to celebrate the Web's birth say he played a crucial role in its development, and they expressed bitterness that his vision had been so discredited.

Gore had been instrumental in introducing legislation, beginning in 1988, to finance what he originally called a "national data highway."

[...]

"He is a hero in this field," said Lawrence H. Landweber, a computer scientist at the University of Wisconsin who in 1980 made the pioneering decision to use the basic TCP/IP Internet protocol Looks like we have to update our Al Gore jokes.

Re:USA! USA! USA! USA!

[QuickFox]

I know those europeons have invented something useful recently but what that may be I have not seen. How about the World Wide Web?

Re: Let's not forget also the TCP/IP unfixed flaws

[myvirtualid]

1- keepalive : IBM's SNA has an efficient keepalive mechanism. TCP has one but I never saw it working properly.

Uh, no? Please do correct me if I'm wrong, but keepalives are not at all part of the TCP/IP "church". In fact, keepalives are outside the TCP state machine, and I've seen some really interesting behaviors caused by poor implementations of this poor idea.

Poor idea? Well, yes: One of the points of TCP is to be able to survive temporary or intermittent "disappearance" of the underlying network. Keepalives expose these failures, breaking the TCP model.

For more on keepalive badness, please refer to section 4.2.3.6 of RFC 1122 (HTML); to RFC 2525 (HTML), which documents a number of known problems with the implementation of TCP keepalives; and to http://tcp-impl.grc.nasa.gov/tcp-impl/list/archive/0367.html for more on why TCP keepalives are a bad idea (and not even implemented the same way on all systems!).

Some of the bad behavior I've seen? This one is cool....

Server and client, both on HP-UX 6.5 (yeah, I know...). Both ends have sockets set to use keepalives (remember the old ISODE code? Used keepalives over TCP to mimic an X.25 analog that wasn't necessary in a TCP network! deep sigh). Every morning, recycle the server via cron. But server won't come back up, claims socket is in use.

Problem? Server shuts down, tells kernel to close socket. Kernel on server tells client that it is closing, both drop into wait state. Socket will close when client sends its close message. BUT! During the wait state, one end (doesn't matter which) sends the other an ACK. WTF? A keepalive.

Due to a bug in how the HP-UX 6.5 networking libraries handled the TCP state machine violation (an ACK at this point is illegal), the close sequence is forgotten, and the two machines exchange ACKs until one is rebooted. Servver's port is never released by kernel, therefore server cannot come back up. Until client machine or server machine (doesn't matter which) is rebooted.

Sigh. Two days with a network analyzer and a lot of WTF, this can't be moments. IIRC, they fixed this in 7.0, broke it briefly later, but fixed it for good in 8.0 and above.

And don't even get me started on the fact that TCP keepalives are still the default in OpenSSH, despite known security concerns with having a keepalive outside the security envelope, and therefore injectable by an attacker.

CIX Commercial Internet Exchange more important

[billstewart]

The Commercial Internet Exchange was probably more important. It was originally set up in 1991 as a peering point between PSINet, UUNET and CERFnet, with help from people including John Gilmore and Mitch Kapor, and it had two important principles

• No Government AUP - Until then, it wasn't legal to use the internet except for government-sponsored projects, though in practice that was interpreted very loosely, with lots of deliberate avoidance of knowledge (especially about Usenet, and especially especially about alt.* :-). You could send email by UUCP, but if you wanted to use SMTP, and talk about business, you had to be very careful to make sure your packets didn't hit the Internet anywhere on your bang paths. You could use Fidonet, though that generally required paying per message or paid accounts, but you also couldn't use Amateur Radio packet networks, which also had a no-commercial-use AUP.
• No Settlements - Unlike the Phone Company business, where every phone call included sharing the price of the call between the carriers who handled both ends and anything in the middle, and therefore complicated everything with billing software, CIX was just "connect up, share the interconnect costs, and deal with your billing yourself." While the peering market has evolved since then, and some carriers buy transit from others, and many carriers will only peer with each other if they've got some balance in their traffic ratios, and some carriers _wish_ they could charge the end users on other networks for using "their" network, and some leftover-monopoly networks still try to charge by the KB/MB/GB delivered, the basic pricing approach on the net is flat-rate, without complex billing arrangements affecting applications or connections across networks.