New Study on Graphene and the internet

[theoristo]

This amazing study suggests that the material Graphene may make the internet one hundred times faster,here some links http://www.bath.ac.uk/news/2013/07/12/graphene-internet-speed/
http://prl.aps.org/abstract/PRL/v110/i21/e217406
what do you think of it?

 

[Greg Bernhardt]

I'm up for anything that makes my internet faster. I'm still stuck on about 1.5mb max download.

I would assume a new material would require replacing cable lines. Would they replace even fiber?

 

[russ_watters]

Sensationalism from a bragging university: the improvement is in switching latency, not bandwidth.

 

[collinsmark]

That first link is very misleading, "could dramatically accelerate internet speeds by up to a hundred times." That's just wrong when put in that context.

It may be true that few-layer graphine can reduce switching latencies, which is all fine and good. But that's not the main cause of latency that you and I experience on the Internet.

The main cause of the communication system latency that we experience on the Internet is the speed of light. If you are communicating with a server even only 3 km away (which is really close, Internet wise), it takes that light/message at least 10 microseconds to reach you. Add to that a small amount for the switching delays, which are on the order of 1 or 10 picoseconds or so (1 picosecond is 0.000001 microsecond.) So using the few-lay graphine switches, might reduce that delay to something on the order of a hundred femtoseconds (100 femtoseconds is 0.0000001 microsecond).

As an example, suppose that there are 100 switches in that system between you and the sever 3 km away. Perhaps the new material can speed up the latency from around
10.001 microseconds to around
10.00001 microseconds
As you can see the delay from the speed of light is the big thing here. The switching delay is comparatively negligible.

And that doesn't even account for the practical, algorithmic latency delays of the server itself, and any router in between: the time it takes to receive a request, and then do something useful with it such as fetching data from memory or a hard drive, and then transmitting that.

And that's just latency. As russ_waters implied, throughput is a whole different ball-game than latency, and throughput is a function of bandwidth, not switching speed.

The first article would have been better if it talked about making transistor switch speeds faster, not making the Internet faster.

 

[DevilsAvocado]

As an example, suppose that there are 100 switches in that system between you and the sever 3 km away. Perhaps the new material can speed up the latency from around
10.001 microseconds to around
10.00001 microseconds
As you can see the delay from the speed of light is the big thing here. The switching delay is comparatively negligible.

Huum... I wonder if there is some misunderstanding... it looks like they are talking about optical switches and optoelectronics (as in electrical-to-optical or optical-to-electrical)... which is quite different from switching as in routing... these switches are used to perform logic operations.

The interconnect bottleneck put constraints on integrated circuit performance due to connections between components. Optical interconnect is a way to solve these problems with bandwidth from 10 Gbit/s up to 100 Gbit/s, and IBM has created a prototype (“The Light-Driven Computer”) for optical interconnect using wavelength-division multiplexing, which if successful could lead to the first computer capable of a billion billion computations per second.

I guess if you can make this technology one hundred times faster it would mean something...

 

[collinsmark]

Huum... I wonder if there is some misunderstanding... it looks like they are talking about optical switches and optoelectronics (as in electrical-to-optical or optical-to-electrical)...

Ahh. Yes I think there was some misunderstanding on my part.

If an electrically controlled optical switch has a switch time of around a few picoseconds, it means the maximum bandwidth transmitted is on the order of hundreds of Gbps (hundreds of gigabits per second -- hundreds of billions of bits per second) per channel (each channel can be a different wavelength of light, i.e., a different "color," used in wavelength division multiplexing). Which makes sense, because the maximum throughput right now is about 273 Gbps per channel, if I'm not mistaken.

The new few-layer graphine switches, if we are to believe the article, have a switching time of around 100 femptoseconds. That places the upper limit on the bandwidth to be about 10 Terrabits per second (ten trillion bits per second), per channel.

So it is about bandwidth.

[Edit: Thinking about this more, the larger bandwidth per wavelength division multiplexing (WDM) channel could/would require wider channel spacing. In other words, even though the bandwidth per channel is larger, there are fewer channels possible on a given fiber-optic cable. This negates most of the throughput benefits, when comparing the new and old, both using WDM. The wider bandwidth per channel could add some flexibility, allow relaxation of wavelength tolerances, and perhaps increase throughput a little bit due to fewer overall gaurdbands. So I'm not saying that the new discovery is useless, because it adds a lot to the whole picture by allowing fewer channels on the same fiber, without sacrificing throughput (which is a good thing). But I'm going to go back to my original statement, which is that saying, "could dramatically accelerate internet speeds by up to a hundred times," is misleading, and arguably just wrong.]

 

[DevilsAvocado]

The new few-layer graphine switches, if we are to believe the article, have a switching time of around 100 femptoseconds. That places the upper limit on the bandwidth to be about 10 Terrabits per second (ten trillion bits per second), per channel.

Bingo!

So it is about bandwidth.

Yep, or throughput, no one is seriously claiming 100 x speed of light...

[Edit: Thinking about this more, the larger bandwidth per wavelength division multiplexing (WDM) channel could/would require wider channel spacing. In other words, even though the bandwidth per channel is larger, there are fewer channels possible on a given fiber-optic cable. This negates most of the throughput benefits, when comparing the new and old, both using WDM. The wider bandwidth per channel could add some flexibility, allow relaxation of wavelength tolerances, and perhaps increase throughput a little bit due to fewer overall gaurdbands. So I'm not saying that the new discovery is useless, because it adds a lot to the whole picture by allowing fewer channels on the same fiber, without sacrificing throughput (which is a good thing). But I'm going to go back to my original statement, which is that saying, "could dramatically accelerate internet speeds by up to a hundred times," is misleading, and arguably just wrong.]

Well... while the physical limitations of electrical cable prevent speeds > 10 Gbit/s the limitations of fiber optics have not yet been reached... As you say, for fiber the current record speed is 273 Gbit/s per channel over 165 km, and with 370 WDM channels NEC managed to get an effective speed/throughput of 101 Tbit/s in 2011.

And there’s new material coming, the University of Southampton achieved a throughput of 73 Tbit/s, with the signal traveling at 99.7% the speed of light through a hollow-core photonic crystal fiber.



They don’t write about this, but my personal guess is that faster/better optical (graphene) switches will improve the ability to do WDM over broader bandwidth (10 THz) as WDM is done by use of optical-to-electrical-to-optical translation at the edge of the fiber...

And there is a lot of funky QM stuff going on in optoelectronics/photonics with opto-atomics and polaritonics as emerging fields (with a polariton as mixture of photons and phonons in the range of up to 10 THz).

So, graphene is cool and I wouldn’t be surprised if we get at least 100 times faster (throughput) internet in the future, and then it’s a different problem that most of us are stuck with the “Copper/DSL dinosaur”...