Slashdot Mirror
Neural Net Routers To Speed Up Net
Virek writes "Researchers in the UK have developed a prototype packet router that is potentially much faster than existing digital technology. They are using a combination of a neural network and "diffractive optical components" (er..?). If they are successful in developing a commercial product they predict that this could significantly speed up the net. "
Alright.. here's my rationale for saying that this cant' do layer 3 lookups.. which is hard to determine conclusively, beause we don't have much information to play with.. luckily we've got some good neural nets upstairs to interpolate with..
To route data, you have to look at the header of the IP packet and read the destination address and then compare it to the routing table and make a decision of which interface will move the data to the destination the quickest. I don't see any way for an optical device to do this. To read an IP header, you have to translate the light pulses to electrical signals and then pass it to the processor to make the routing decision.
The way I see this working is it has several streams of date to move from interface to interface, it knows which interface it has to move them to, and this technology is finding the fastest way for the switch to move multiple streams of data across the backplane simulteneously. It would sort of turn the backplane into a totally optical, neural net based ethernet, as it is in a sense a CSMACD (Carrier-Sense Multiple Access Collision Detect), multiaccess media, with the neural net avoiding collisions and finding the best way to transport the data..
of couse.. my neural net might be returning an invalid result for these inputs..
//Phizzy
"Most European technology just isn't worth our stealing," -- Former CIA chief James Woolsey, referring to Echelon
I'm afraid you've not kept up on current technology. A laser diode is about the size of a grain of rice, and uses milliwatts of power. Also, laser diodes aren't very expensive (consider that you can buy a laser pointer for $20). While the laser diode in a laser pointer isn't the same as the lasers they are using in this, it's not THAT much different.
www.eFax.com are spammers
Routers have two components : a routing component and a switching component. The switching component takes input on an interface, moves it across the backplane to the routing component, and then and pushes it out another interface when the routing component is done with it. The routing component reads the destination from the IP header, compares it to it's routing table and decides whether it is routable, and if so, which interface it will go out. It seems to me that this technology would only take care of the level 2 (switching) layer of networking, which is not the major area of concern in networking right now.. With existing electrcal technology, we have backplanes that can push 160Gb/s of IP traffic. That, on one device, is enough to push 16 OC-192s. The problem is if a single device had to push 16 OC-192s at wire-rate (full speed, basically), the route processor would not be able to handle the IP lookups to feed the backplane. So this technology (if they can get it to work.. which it seems like they should be able to) _will_ be useful once we get a route processor that can keep up with it.
This tech looks like it could be separated from a router to be used in an ATM switch, but by the time it is viable, we probably will be using MPLS and POS more than ATM, but there will always be a need for fast layer 2 technologies.
I like the current trend in networking to move away from electrical processes in switching.. once we can get a fully fibre-based system we will be able to remove much of the inherent latency in telecommunications.. like the 100ms lag to go across the atlantic, or the 60ms lag to go across the United States. Once we have comparatively few electrical devices taking care of the routing and all optical devices doing the switching, the only thing that will cause lag will be processing and the speed of light.
//Phizzy
"Most European technology just isn't worth our stealing," -- Former CIA chief James Woolsey, referring to Echelon
AC, let me introduce you to something called a "mirror"...
The neural technique is simply a means of choosing the optimum non-blocking configuration. Go and read a textbook on comms theory and come back when you understand why an M x N non-blocking switch is a hard problem.
I'm not going to read a textbook to understand an article such as this. And that's the key flaw of the article: I do have solid background knowledge in computers and networks, but I'm not a specialist in that particular field. Now, it may be correct that neural networks are a good thing in that context, but the article totally failed to explain anything.Seriously, I'm a CS major, and I did attend the odd lecture about neural networks. From what I've heard there, neural networks are a tool to implement rules the exact details of which you can't specify very well, i.e. image recognition. I fail to see, on a very basic level, why they should be better than a good, solid priority queue for traffic scheduling.
The illegal we do immediately. The unconstitutional takes a little longer.
--Henry Kissinger
Of course the bloody thing is fast --- it uses optics instead of old fashioned electrons. If I were to build an idiot-simple optic router that uses a routing scheme from the 1970s, I'd be willing to bet it would be pretty fast too!
Stories like this remind me of one of my favorite papers from the Annals of Improbable Research. The Paper is titled "Advances in Artificial Intellegence since the 1970s" (or something like that). it has three German co-authors, and consists of a blank page:)
...but I would prefer having smarter users. Ones that have heard of the concept of "local mirrors", for example.
For around a year now, I've wondered if there's any advantage to mimicking the visual cortex in networking. To me, it seems like the bandwidth would be immense. However, the range is significantly short with this sort of connection.
Sci-Fi always contained stories of people with cybernetic parts fused to them. How about the other way around? It would seem we're a more efficient computer than the turing model.
This is a great idea. It's hight time the internet moved forward with technologies like this, but the real question is this: are we ending up with too many new technologies, and as a result the demand for qualified individuals - individuals that understand the fundamentals of routing and internetworking and can learn these new technologies and integrate them with current technology, becoming too much - ie, the demand for them outpaces the rate at which individuals that come along with this knowledge become available? I think so, and, working in the industry, I've seen it happen all too often: the Internet in general is facing a problem: As a Slashdotter pointed out in the comments section of the story about the great Internet pioneer, Donald Davies, there are fewer and fewer people who understand Internet fundamentals fully. Yes, there are a few - great hackers, CCIE's, and old-timers. But new IT professionals walk in with their CS degree or MCSE and really don't know much. They probably couldn't even tell you what a /30 or a /29 is. More CCIE's are the solution, some people would say, but the fact is that the CCIE qualification, while good, ties people down to one vendor - Cisco, which will create a kind of Microsoftian situation, except with networking equipment and protocols (ever hear of the Cisco Properietry protocols? Ayup.)
However, a solution to this problem is at hand. The GNU/Zebra. This is a robust routing daemon which is fully open sourced, released under the excellent GNU license. I can't remember the URL offhand, but there is a large project which aims to implement a CCIE-like course based on the product, and which also includes a lot of GNU/Linux material. The course will take the candidate through Linux basics, and then move on to Networking basics, and then advanced Networking (BGP, IGRP, OSPF, etc). They will then be taught how to apply these concepts with the GNU/Zebra. The reasoning behind this is that long after Cisco has died out, Zebra and Linux will live on, due to their open source nature. It's a common argument against all propreirty companies, and , I think, a valid one. Even if the company is huge, it might not survive in all sectors, and if it is a hegu one concetrating on one sector - it might go down completely with advances in technology and demand for flexible solutions that the company, no matter how big it was or how it tried to cover al l the bases, so to speak, failed to see and got left behind. This isn't unlikely to happen in the Networking arena, and Cisco is a prime target for the kind of failure described above. The ZIE (Zebra Internetworking Expert) course will also be amied at educating more people, even though it will be fairly expensive at first, it will be priced reasonably once it reaches a certain level of acceptance by the business community. Their will be 3 exams. 2 written, and one lab. The 2 written will consist of Linux basics and Networking, (which will include both Networking basics, in the Linux module, intermediate Networking and advanced Networking in the Networking module). In the lab, the candidates will be asked to simulate a real working eviroment and interface with other hardware and software, using the GNU/Zebra technology.Availability: These exams will be available through various LUGs around the world, eventually. At first they will probably only be available through a few centers in major areas in the US and Europe.
Pricing: There are several factors in the price here, and although it will be more expensive overall than the CCIE at first, the prices will go down as the qualification gains acceptance by the business community and the availability goes up with the rise of Linux, Open Source, and cooperating from various LUGs and other Linux/Open Source organizations. The written exams will be around $250 each, with the lab exam costing $1,000. If the candidates fail their first time, they will be given a compensation period of 2 months in which they can rewrite the lab exam for $500. The most expensive part of the course will be travelling to Zimbabwe. It is neccessary for all candidates who have completed the lab and written exams to be initiated in Zimbabwe, where they will need to have sexual intercourse with a Zebra mare in heat. At present, female candidates aren't provided for. This will change as the conditions above are fulfilled , though. The trip itself will cost $5,000, hotel accomodation included.It's the hope of the organizers of this qualification that it will encourage the use of non-propreitry solutions for networking in mainstream organizations, and promote the use of Open Source technology in general. It's believed by them that the Open Source methodology will not only lead to technical benefits, but will allow communications to go to the next level by bringing back the cooperation of old that started the internet and allowed it to grow, in short: the hacker and scientest and military cooperation of the 60s, 70s and 80s., without which such innovation would have been impossible
I made it onto the anti-spammers list with one of my accounts. Do I get a prize?
--
Are you sharing whatever it is you're smoking? I think I'd like some....
First, make it work, then make it right, then make it fast, then, make it bloated!
Just one more step in the Internet's quest to become a sentient entity and destroy all the meatheads. I can't wait. Oh... wait...
I'm trying to teach myself to set people on fire with my mind... Is it hot in here?
"I like to sit and cry in front of my T.V." -- T.V. Song - Moondog Jr.
I am sure this is some kind of hoax. Just ask Minsky. Neural networks don't and can't do anything. They are a dead end, and research must be supressed as Perceptron research was before them. You foolish mortals have no idea what a dangerous^H^H^H^H^H^H^H^H^H silly idea neural networks are. After all, if it trains itself, how can I credit my genius for its success?
(Humor)
Possibly because you're never going to be able to implement a priority queue which will be effective for all kinds of traffic w/o having some kind of dynamic adaptation to the changing traffic characteristics?
One thing that distinguishes neural networks from other approaches is that the "cleverness" (ie. the effective rulebase that determines what happens) is embedded implicitly in the internal state of the network as a set of trigger weightings, and the pattern of weightings has absolutely *ZERO* obvious relationship with the work being performed in any sense that is meaningful to humans.
It's bad enough trying to diagnose routing problems right now, in a world of dynamic but still fairly deterministic routing algorithms. Add neural networks into the equation and all bets are off: we'll never be able to determine why any particular route was taken. (An explanation of "because those 57 nodes fired" is not particularly helpful).
Oh joy.
"The question of whether machines can think is no more interesting than [] whether submarines can swim" - Dijkstra
Uh... What's so much more dangerous about Hydrogen in comparison with gasoline??
The illegal we do immediately. The unconstitutional takes a little longer.
--Henry Kissinger