Slashdot Mirror
DARPA Aims to Redo the Internet Protocol
Xaleth Nuada writes "The Defense Advanced Research Projects Agency's (DARPA) is looking to redo the entire Internet Protocol. With the DoD increasingly adopting network-centric warfare the shortcomings in the current IP have become resoundingly clear. Everything works fine for static hardwired networks. But not for dynamic wireless ones. The benefits for your average geek? How about REAL wireless networking? Easier network set-up? Increased wireless security protocol? Increased reliability in sending information?" Don't forget massive incompatibility and upgrade hassles. :)
Given the scale of the re-work proposals (replacing the Von-Nuemann architecture...), I'd be surprised if there wasn't some effort made to embed snooping and tracing into all packets transmitted. This *is* the DoD after all!
On the other hand, given how slowly IPv6 is making its way into the wider world, we probably don't have too much to worry about for the time being!
Simon
Physicists get Hadrons!
They blame the packet nature of the network for lots of the problems but I see not other perposal given. How on earth do you build a network as large as the internet based on a non-packet archetecture? I am studing computer science right now at school and haveing completed two telcom courses and nobody has ever discused a conection-oriented technology that or even a conection-oriented concept that could cope with a network as large as the internet with as many hosts. Do any of you in slashdot land have a clue how they might even start to go about doing this? The other posibility is its a new twist on a conectionless network but how on earht is that possible with out some sort of packet archetecture to send over it, otherwise you'd have no way to change path with conditions and changeing conditions are UNAVOIDABLE on any network I have ever seen.
Repeal the 17th Amendment TODAY! Also Please Read http://www.gnu.org/philosophy/right-to-read.html
Obviously the writer of the article and Gibson don't understand how the system works at all... they're with the normal public thinking that e-mail is being transfered from place to place as some whole document and not understanding the basics of packets or anything in TCP/IP.
I am not a network engineer... but I am pretty sure that if you wanted to assure the delivery of email you would do it at a HIGH level in the stack, not at the transport level. If they are talking about packets, it has already been done. I am not sure that the Gibson in the article really understands what he wants.
It's pretty clear they've got the ideas and concepts all screwed up here.
Why don't you embrace your slashbotness instead of living in a dreamworld?
It sounds to me more like some general had a brief introduction to computing theory, but didn't relate it to any real current technology.
The alternative to Von Neuman (Code and Data in the same memory) is to have code and data in seperate memory areas. This makes it very difficult to make computers where the code can change. Sure, there's no buffer overflows, but there's no security patches either. It might be fine for embedded devices, but I'll not have it on my desktop. The Page (or Segment) executable flag of more modern memory management units does the job fine, without all the hassle.
The OSI model is already not used anywhere except to compare proposed network models to; it's way too complex.
He talks about replacing packet switching so that messages are delivered on time & with certainty. Presumably he means some kind of virtual circuit switching, but he also talks a lot about constantly shifting ad-hoc networks. Circuit switchinfg & ad-hoc networks don't mix well. You have to know what the path is going to be before you can reserve it. It's probably better to just turn on the QoS and AH already implemented in IPv6.
"seconded. I don't see what is the problem with von Neumann architecture, and the article is pretty vague about that."
e ) try to fix these problems. (And I'm guessing that strict seperation of code and data might ease formal proof?)
The von Neumann archicture doesn't distinguish between instructions and data, allowing a program to modify another program or itself. (Think viruses/trojans.) But I think memory protection has patched this pretty well.
It also has a memory bottleneck. Other models, such as Harvard, (http://en.wikipedia.org/wiki/Harvard_architectur
I don't know of any great solution to the problem of starving the processor with slow memory access etc. but I think this is where you would look for one...
--Mike---
Flaws in the basic building blocks of networking and computer science... "It is time to ask the harder questions about the ways of computer architecture we've been using for the past 30 years. Is it time to scrap the von Neumann architecture?"
This is the only interesting part of the article. I couldn't care less what they do with the OSI layers. As long as someone writes about it as well as Stevens wrote about TCP/IP, it'll take me a month of reading and programming to get under my belt. We all learned Pascal, then C++, then C++ again when the standard came out, then Java, and Lisp, and Smalltalk, and Perl, andd Python, and C#, and a half-dozen more languages as the need came up. Now, you have to learn a few new networking layers and protocols. No big deal -- you should be pretty damned familiar with learning different implementations of stuff you already understand.
But, replacing the von Neumann architecture means changing just about everything I know. That's big. Everything is von Neumann. All the computational models, all the theory, all the basic underpinnings of what I know... it's all pretty much out the window once von Neumann goes. It's not just a dozen evenings at home with a book and reference implementation to relearn all of that stuff, either. It's relearning nearly all the Computer Science I know, and probably learning a whole bunch of new Maths to go with it.
That's gonna hurt.
Slashdot is jumping the shark. I'm just driving the boat.
The scary thing is, the underlying concept there is actually plausible. Think about the similarity between human social connections and the connections between neurons in the brain. You're not aware of being part of a collective consciousness called humanity, but the individual cells in your head aren't aware of being part of a larger consciousness either.
You have to wonder how many things we consider "miracles" or extreme luck could really be actions of a larger entity which can influence groups of people as effortlessly as you can flex your fingers.
Map the cells in the state tables to appear as conventional RAM to the host, and reprogramming becomes as easy as a memory write. Bad cell?, just route around it. The fact that it's all state driven allows you to build an automated rerouter almost trivially.
post Von Neuman computers are going to be wicked fast, if they can build IO to keep up with them.
--Mike--