Internet Heading to Light Speed

dbaker writes "Wired Magazine has a very interesting article about the future of optical networks and the barriers we face before this technology is commonplace."

Comments

A "light" transistor to the rescue!

[erick99]

The buckyball layer produces what sounds like a transistor of sorts:

The gluing process creates a material composed of larger electron-rich molecules with sufficient power to cause light that passes through to control the direction of other light, providing the switching capability, Sargent said.

With switching occuring at the speeds available through a layer such as that, there would be an incredible decrease in cumulative latency across the 'net. That is, if all or most of the switches are as above.

Superconnect's Lehenbauer agrees that "it's fascinating" to have material for an optical switch, but warns "it could be awhile until an all-optical network is possible."

I wonder what the cost of those type "devices" will be - both direct in terms of the devices and indirect in terms of whatever infrastructure is required to implement them. Well, either way, it's great sounding technology.

Cheers,

Erick

Re:A "light" transistor to the rescue!

[TrumpetPower!]

"Light" transistor, indeed!

One of my first thoughts upon reading the article was that that's exactly what they've created--an optical transistor.

It gets even better. The original transistor originally played a huge role in replacing human operators in telephone network switches. That also seems to be the first target for this new breed of transistor.

Surely, the optical computer just became much more of a possibility. Yeah, we're still a long way from an optical IC, but this is a big step on that path.

Cheers,

b&

Re:A "light" transistor to the rescue!

[Noctambulus]

The gluing process creates a material composed of larger electron-rich molecules with sufficient power to cause light that passes through to control the direction of other light, providing the switching capability, Sargent said.

With switching occuring at the speeds available through a layer such as that, there would be an incredible decrease in cumulative latency across the 'net. That is, if all or most of the switches are as above.

Although I am by no means a router expert, it would seem logical that a majority of the latency in the network is caused by the actual reception and subsequent "analysis" of the packets. After all, the "response time", or "spped" of electricity is at best close to the speed of light.

Thus, a majority of the time spent when moving packets around is probably spent on the routers, when processing the information. As such, switching to light based media should not have a noticable effect latency.

However, with the change of media, more data can probably be sent through the pipe, therefore increasing the overall speed of transfer in a favourable way.

Regardless, the technology certainly sounds exciting.

Re:A "light" transistor to the rescue!

[Marxist Hacker 42]

What I don't get- couldn't you accomplish the same thing by encoding the addressing into the color layer and just using a prisim for a switch?

Planning for the Future

[The Subliminal Kid]

Redundancy may soon be more vital than speed.

Once the internet was designed to withstand problems (a euphemism for a nuclear strike) at multiple nodes but since commercial interest like to keep as many things as possible in one building we see today that a small fire in a maintenance tunnel has a dramatic effect on the over all network latency. There just isn't as much redundancy as there used to be and that may be worse for us all than your download time for SP2.

Re:Planning for the Future

[dwdm_dude]

Of course, enterprises with high bandwidth needs have typically had to pay through the nose to lease capacity (redundant or not) from the huge phone conglomerates. Now they can reasonably buy their own 100Gbps+ equipment to connect data centers and even justify redundant links due to the cost effectiveness of new enterprise-friendly applications of tried-and-true carrier-grade technology. Celion Networks, for example, ships such systems.

Sounds great, but...

[Anonymous Coward]

What happens when you put all the switches together and actually have to route the packets, and the next hop is "busy" on that light frequency already?

You would either have to:

a) shift the frequency to a different portion of the light spectrum, or

b) somehow delay the light signal until the previous message is completely transmitted through the router.

But without using a light-electrical-light conversion?

I don't know how a) could be accomplished other than using one laser to pump another (but there would not be enough intensity for that), and using cryogenic sodium to slow the light pulse down long enough is not practical in a low cost router (yet).

Any ideas? Or did I miss something obvious here?

Optical routers already exist

[cayblood]

Bell Labs invented them in 1999.

"You do not need *light* to get *speed of light*!

[PaulBu]

... as my former advisor Prof K. Likharev used to say. When you send a sharp electrical pulse down a matched transmission line/waveguide it propagates with, you guessed it, speed of light in the medium. If your insulator is the same SiO2 they use for optical fiber you will get the same speed as in the fiber!

The problem with traditional voltage-based electronics at 40G speeds is that when you drive a SiGi/InP/GaAs transistor that fast it dissipates LOTS of power (measured in Watts per handful of transistors). Moreover, CV^2f/2 power dissipation when you constantly charge/discharge line capacitance to ~1V operating voltage is significant. And of course the maximum operating speed of any substantial logic is determined not by transistor speed but by RC constants of the wiring.

Now, if one departs from traditional transistor logic design, say, to superconductor electronics (which I've spent all my life designing up until the beginning of this year, when my current employer decided to "discontinue that effort"), you can start from a clear sheet of paper. In superconductor case, first of all you lose R in RC, not bad! Second is that when temperatures are that cold, thermal noise (~kBT) is small and operating voltages (pulse amplitudes in our case) could be ~1 mV, not ~1V, and Josephson junctions are pretty happy generating ~1ps wide pulses.

The downside is having to deal with refrigiration, one would not see this technology on the end user's desktop any time soon, but for the telco switching center it is almost doable.

My personal estimates (well, down to the complete circuit diagrams ;-) ) showed that we could make a 128x128 non-blocking self-routing packet switching matrix at 60Gbps/line that could fit on the palm of one's hand, and after packaging with refrigerator fit on half a rack.

The "packet" feature is important, often when "optical computing" people talk about their switches they conveniently omit the fact that while switch might be fast enough for some 120GHz of bandwidth the re-configuration of that takes milliseconds (think long-haul traditional SONET lines), we were talking about routing/re-configuration at ~256 bits packet length (think TCP/IP).

Oh, well, it's a pity that I can not work on this stuff now, it was -> |- THIS close to actually coming up with a viable demo/product. Maybe some day...

Paul B.

pbunyk (at) lycos (dot) com

P.S. Google for SFQ/RSFQ for more info

Re:magnetic media

[dmayle]

The joke's on you. Besides being thought of before (as mentioned by another poster), it's how Cray used to store memory for their super computers. There was a wire that went from the output to the input, and the clock was timed to match the duration of the electron transfer through that wire...