Log in

No account? Create an account

Previous | Next

On "magical" technology

The University recently spent quite a lot of money buying four strands of "dark" fiber optic cable in a vague ring shape around the state, connecting the three campuses in Chicago, Urbana, and Springfield. To put actual networking services on this cable, you need something that can a) optically drive that much fiber so that the light signal is recoverable at the far end, in some cases hundreds of kilometers away; and b) can make best economic use of those very few strands of very expensive fiber.

The solution is a technology called Dense Wavelength-division Multiplexing or DWDM. This allows us to pack up to 40 channels, of up to 10 Gb/s each, onto a single fiber strand, by sending each one on a slightly different wavelength of light. Because the window of transparency of optical fiber is fairly narrow, the different wavelengths or DWDM channels are packed in pretty tightly, less than a nanometer apart. The physics involved in making all this work just boggles my mind.

You need something to break apart the combined signal with all channels on it into separate signals, one for each channel. Since this is just a matter of picking out different wavelengths of light, this is no more mysterious than a prism or diffraction grating. However, picking out a VERY FAINT optical signal out of an optical fiber, sending it through a diffraction grating, and then jamming the VERY PRECISE AND PACKED TOGETHER individual signals each back into their own optical fibers seems pretty amazing.

You need a way to reamplify or boost the signal every 50 kilometers or so. The economical way to do this is to figure out a way to boost the whole thing, all 40 channels at once, on the fly. This is done with something called an Erbium-doped Fiber Amplifier or EDFA. A "pump laser" raises the electrons in atoms of the Erbium dopant to a higher energy state; the signal photons flying down the fiber then interact with these energized atoms, stimulating them to emit more photons and thus amplifying the signal.

Finally, different wavelengths of light propagate through glass at different speeds (this is why a prism can break white light up into a spectrum). The narrowband pulses may be less than a nanometer in width, but this is still finite, so this chromatic dispersion will, over enough length of fiber, tend to smear out those pulses into miniature spectra. Eventually the pulses run into each other and the original signal is lost. To compensate for this, special glass is made with a chromatic dispersion coefficient opposite that of the "travel" fiber. So after traveling and smearing out over 70km of fiber or so, the signal is amplified and then sent into "dispersion compenstion fiber", 10km or so of it wound neatly on a spool. This reverses the smearing effect and allows the slow wavelengths to catch up with the faster ones, neatly and magically undoing the dispersion effect.

It's all magic. I can explain it, but to understand it fully requires math and physics that are way beyond me.

Several of us spent four days at Movaz, the company that makes the DWDM equipment we will be using on our dark fiber network. It was incredibly useful and incredibly information-packed. So much so that I felt a little dizzy and spaced out at the end of every day. But now I know an awful lot about their product and DWDM theory, and am confident that, even if I don't have to make this stuff WORK, that I can fix it if it breaks, and exploit it to its fullest potential.

We got to work on live equipment in their lab, and it was a fucking blast. I took a few pictures with my phone. Click for Gallery.


( 11 comments — Comment )
Sep. 10th, 2006 08:19 pm (UTC)
A big gun in the UCSD Engineering college gave a seminar at Qualcomm about this technology and covered most of the points you mentioned. His context was trans-oceanic fiber cables and how the erbium-doped amplifiers had given an order of magnitude increase in their reliability. It's a very expensive proposition to send a submersible to the bottom to fetch up a cable for repair.
Sep. 10th, 2006 08:31 pm (UTC)
Apparently 10 Gb/s signals are really pushing this stuff as it is. There are systems that can go to 40 Gb/s, but at those rates polarization-mode dispersion (due to different orientations of the EM field propagating at different rates because the fiber isn't perfectly round) becomes important, and there's not a good way to compensate for that.

The 100 Gb/s Ethernet standard is nearing the point where vendors will start producing stuff. I have no idea how THAT stuff is going to get transported.

The existing technology is pretty mature, though: I marveled at holding an OLD in my hand. It's about the size of two cigarette packs lined up end to end, and has a 30dB EDFA, MEMS-based variable optical attenuators, filters and combiners to drop and add the 1310nm control channel, and a full complement of instrumentation and diagnostics. Two boards mounted face to face, with the optics on one and the electronics on the other. And apparently reliable as can be (meets NEBS standards for MTBF etc), and no muss or fuss with interchangeable parts: if it fails, just slap in a spare.
Sep. 11th, 2006 02:31 pm (UTC)
Why do you need expensive equipment? We found cheap info tubes for the internets at a superstore over the weekend. I bet you could buy them in bulk! You can't put a horse through this kind of internets but how often do we really study horses on this campus?

The machines are pretty :)
Sep. 11th, 2006 04:05 pm (UTC)
Oh! and...
Last week we saw a USB turntable. *sparkly eyes*
Sep. 11th, 2006 04:32 pm (UTC)
Re: Oh! and...
So you can control what 33⅓ LP's get played by clicking your mouse on your computer? How quaint!
Sep. 11th, 2006 04:35 pm (UTC)
Re: Oh! and...
You can also rip directly from LP. And I have literally hundreds of old Monkees, Rod Stewart and Chipmunks albums waiting to be shared with my closest friends.
Sep. 11th, 2006 04:29 pm (UTC)
They really were. I think impressive equipment that has that modest, handsome look to it is just really sexy.

We should tell Senator Ted Stevens that the Internet these days is not like a set of pipes, but really more like prisms and rainbows. That'd really confuse him.
Sep. 11th, 2006 04:34 pm (UTC)
Sep. 12th, 2006 05:58 am (UTC)
We use 2 pairs of Nortel Optera 5500s on dark fibres between our 2 datacentres, but those fibres are only about 60km long.

But the shiny part is the amount of crap they can cram down each fibre.

Each optera will take I think 6 or 8 interface cards, each card will then have 10 LC MM fibre pairs on it, you can configure those either as fibre channel or GigE.

So blooming simple, just connect your switch/routers existing Gbic to the optera and off you go.
Sep. 26th, 2006 04:50 am (UTC)
Heh heh, I'm not intimately familiar with the Optera gear (they bid on our equipment solicitation but were not favored), but we just got delivered to us a new gigabit Ethernet ISP service from AT&T. Rooting around in the equipment racks to see where they delivered our gig circuit to, we found it coming out of... an Optera 5500.
Sep. 12th, 2006 08:52 pm (UTC)
Wow, that's incredibly cool. Thanks for explaining.
( 11 comments — Comment )



Latest Month

July 2013