100 Gbps Via Ethernet

Doc Ruby writes, "As reported at GigaOM, 'Infinera has bonded 10 parallel 10 Gb/s channels into one logical flow while maintaining packet ordering at the receiver,' bridging 100-Gbps ethernet over 10 10-Gbps optical WAN links. Infinera's press release is here. Further from GigaOM: 'The experimental system was set up between Tampa, Florida and Houston, Texas, and back again. A 100 GbE signal was spliced into ten 10 Gb/s streams using an Infinera-proposed specification for 100GbE across multiple links. The splicing of the signal is based on a packet-reordering algorithm developed at [UC] Santa Cruz. This algorithm preserves packet order even as individual flows are striped across multiple wavelengths.' We're all going to want our share of these 100Gbps networks. The current network retailers, mainly cable and DSL dealers, still haven't brought even 10Mbps to most homes, though they're now bringing fiber to the premises to some rich/lucky customers. Are they laying fiber that will bring them to Tbps, or will that stuff clog the way to getting these speeds ourselves?" Rumors say that what runs over Verizon's FiOS is ATM, to support their aspirations for triple-play.

Re:Rich?

[Jim_Maryland]

Recently switched to FIOS myself and I'm saving compared to my Comcast bill.

$117 Comcast (Digital Cable without any of the premium channels, broadband at 7 Mbps (although never actually saw rates near that))
$69 Verizon FIOS (Digital TV without any of the premium channels, broadband at 5 Mbps (actually saw rates exceeding that but generally very close to the advertised speed))

Since Verizon FIOS was available in the area (Maryland), Comcast has been pretty heavily advertising their bundle for new customers where you get the Digital Cable ($33), VoIP ($33), and Broadband ($33). Unfortunately that appears to be a one year deal compared to Verizon FIOS which doesn't appear to have plans to jump up after the initial year (hopefully I didn't miss some small print).

FYI - Using SunRocket [sunrocket.com] as my VoIP with the monthly cost under $17 so the Comcast package isn't an option for me.

TCP-PR = neat stuff

[neuro.slug]

Here's a link to the paper (PDF) [ucsc.edu] on the packet reordering if you're interested. Being a former banana slug, I was very interested to see this research coming out of UCSC. It makes me feel warm and fuzzy whenever something theory-based is actually implemented.

Certainly welcome in the data centre world

[anticypher]

This is a natural progression of ethernet speeds. 10GigE switches are getting to the price point now that we are installing them everywhere. I even had a 10GigE switch on my home fibre for a week of testing, but slashdot just doesn't load any faster.

All the broadband providers are moving to larger pipes now, with GPON (Gigabit Passive Optical Network) going in everywhere, as consumers are clamoring for more than ADSL2+ speeds (24Mbps down, 2Mbps up) in city centres. I'm designing the back end of a GPON network, where every neighborhood gets 2.5Gig down, 1 Gig up, shared between 16 residences. Of course, there is going to be more than just internet on pipes that big, quadruple play to start, and as new services become available even more bandwidth will be needed. Once you start piling up the 10GigE connections, it will be nice to have a working trunk/etherchannel/bonding solution for those long hauls between data centres.

the AC

Re:Whoa.

[ebob9]

Well, thats closer, but I think in reality it would be a bit smaller.

So, you did 10 (Gbit/sec) * 8 (bits per byte) * 5 (seconds) = 62.5 Gbytes.

The 10Gb links are Ethernet links. Lets also assume HTTP is the transfer method, just to make it 'easy'. We could use FTP which is UDP but then we'd have to account for the TCP Control connection in the traffic. Heck, Lets even assume nice jumbo frames with a 9000 MTU. Also, lets assume the video is 'optimally' compressed.

Ethernet header = 14 bytes
IP header = 20 bytes
TCP Header = 20 bytes

14+20+20 = 54 bytes out of every 9000 transfered for header information.

On top of this, there will be HTTP headers at the start of the request, but since they are only transfered at the start (not every packet), lets factor them out as miniscule.

So, basically 62.5 is the maximum theoretical data of the circuits. 62.5Gbytes /9000 * 54 = 375Mbytes of packet overhead.

The Maximum possible transfered in 5 seconds would be 62.5Gbytes - 375Mbytes = 62.125Gbytes.

FTP uses UDP?

[yabos]

Not according to the almighty wikipedia
http://en.wikipedia.org/wiki/Ftp [wikipedia.org]

Re:Bad idea for the home network...

[QuantumRiff]

It is one cable. They split the signal out to the single fiber cable using different wavelenghts. (ie, colors) and run 10 wavelengths over the same cable, at 10Gbs each. For simplicity sake, picture it as 10Gb/s with a red laser, 10Gb/s with a Blue laser, 10Gb/s with a green one... and so on.

what's so special (besides the speed)?

[Creepy]

The idea of packet order guarantee and ethernet are pretty much mutually exclusive, and upon further review (reading the fa), it reassembles out-of-order packets pretty much just like TCP. About the only new thing is the out-of-order disassembly and assembly and the overall speed. It still has the same flaws as Ethernet, which is that it really is only about 80-85% efficient - after that you would be better off with a Token Ring or other managed protocol (token rings are excellent for saturated networks but poor for low saturated networks).

Sigh - and the poster seems to think ATM is a good protocol, but ATM is a terrible protocol, especially for data, but even for voice it's mediocre. It was designed for voice conversations over high noise lines with significant data loss (copper) and predominantly used over low noise high speed lines with almost no loss (fiber). Its advantage is standard packet length (53 bytes) and speed. Worst disadvantage - almost 10% overhead (5 bytes of every 53, or ~9.4%). ATM also has no guarantee of sort order or collision avoidance (since it's asynchronous) so in practices it can be really bad. Incidentally, my networking class voted this the worst protocol back in 1996, but expected it to succeed mainly because of telecoms pushing it.

Re:Why stop there?

[TeknoHog]

It's all about bandwidth, which is not the same thing as date rate. Unfortunately there's a huge marketroid conspiracy trying to teach us otherwise.

Data rate (in bps) is proportional to bandwidth (in Hz). The factor between them depends on the modulation and coding schemes, which in turn are limited by the signal/noise ratio of the medium. Anyway, the system of light sources, fiber, and receivers has a certain limited bandwidth. For example, if you're using visible light from about 400 to 750 THz, you have a bandwidth of 350 THz, assuming the fiber works across the whole spectrum.

For a naive example, imagine that it takes 35 THz of bandwidth to achieve 10 Gbps over a certain fiber. Then you can have a maximum of 10 of these channels using visible light, if you shift the center (carrier) frequencies accordingly (i.e. different colors) to avoid channel overlap.

For another explanation, imagine using a single frequency of light to send information. As soon as you start switching it on and off, you're introducing other frequencies; the spectrum gets wider, the faster you switch (see also: Fourier transform). So, if you try to use zillion distinct colors to get zillion channels, they will start to overlap at some point if you try to send any information.

Re:FIOS in Maryland

[ScentCone]

So, while I don't know if Verizon is bundling the service to consumers, Verizon is certainly bundling the services to the Counties, since Verizon already has whatever permissions it needs for internet and voice services.

My Mom (in Montgomery County) has FIOS up to her house for data. It's fast. She's using Comcast for cable TV. She called Verizon about switching over to them for 'cable' service so she could bundle that, and her copper land line all together. They said "Sure! we'll have someone out to talk to you about that tomorrow morning!" She asked me to stop by just in case there were going to be "any of those techie issues."

Well, there was. The "issue" was that Verizon's cable service was actually an affiliate sale of DishNetwork (or was it DirecTV?), and there on the lawn was a guy ready to bolt a dish to the side of her house. And it was a 2-year commitment! I'll allow for 10% of Mom not knowing what she was doing on the phone with Verizon, and 90% Extremely Sleazy Customer Service Tactics. Incredible. So, with the guy standing there, screwgun in hand, I called Verizon, who told be that it was "too late" to cancel the "order" but that in 24 months, we could switch over to FIOS TV, which would surely be available by then. We told them that it was also too late for the lawyer's phone to stop ringing, and presto the installer (who spoke no English!) was getting a cell call, and he had the dish packed back up and in his van before we got into our second cup of coffee. Beware the cable/FIOS state of flux in Maryland, especially Montgomery County. Of course, you also have RCN as another option.