Good News From The High-Speed Networking Front

Degrees writes "Over at Small Times there is an article about two Danish companies that want to make deploying fiber optic lines easier with MEMS-based packaging technology. (MEMS is Micro-Electro-Mechanical Systems - described here). Also mentioned is that the big three U.S. telcos are working on fiber to the home plans." And punkmac points to this eWeek article which begins "An Intel Corp. backed startup, SolarFlare Communications Inc. said Monday that it has developed a working prototype of a chip that will permit 10G-bps communications over standard CAT5e copper wiring. SolarFlare's chip will be used as evidence that 10G-bit over copper can be done, in anticipation of a draft IEEE standard to be developed later this year."

Comments

10Gbps over Cat5e

[Ray Radlein]

Obviously, this ain't coming to the home for a few more years (heck, Gigabit switches are only just now getting home-use priced), but it'll sure be nice to not have to re-pull all that Cat5e cabling we ran all over our house, especially since we'll probably be in our fifties by then.

At that type of transfer speed, the network should effectively vanish completely, even if we're streaming HD video to or from the downstairs entertainment center (I'm assuming that the internal bus bandwidths in the computers will have improved proportionally as well by then).

Re:10Gbps over Cat5e

[Ungrounded Lightning]

Anyone know what the theoretical speed limit of copper cable is? 10Gbs seems faster than copper can go to me.

Depends on its length, thickness, surrounding dilectric, shieling/balance/discontinuities, and the speed of the carrier/modulation. (For any given design of wire it's mainly the length.)

Copper, not being a superconductor, has resistance. The resistance combines with the stray capacatance between the conductors to form a distributed RC low-pass filter/delay line, which attenuates and delays higher frequencies more than lower frequencies - progressively more as the wire gets longer.

It gets even worse for REALLY high frequencies, because they create eddy currents in the copper that impede the penetration of current into the conductor, restricting the current to the outer part of the conductor (the "skin effect") and thus raising the effective resistance and exaggerating the frequency-selective attenuation.

This selective attenuation and delay weakens the signal - more at high frequencies than at low. As the wire gets longer the signal gets weaker and competing noise pickup gets stronger, reducing the signal-to-noise ratio and thus the amount of signal that can be carried.

But the selective attenuation and delay also distorts the waveform, creating "intersymbol interference" (stored charge from previous bits affecting the latest bit). This can be compensated for.

Current technology using SERDESes (fast serial bit streams), with some compensation for the selective attenuation (both preemphasis at the transmitter and compensation at the receiver), can get 3 Gbps through about a yard of printed circuit, or several yards of wire. More advanced devices (using tricks like four-level encoding to get two bits per modulation perios and feedback from the receiver to the transmitter by a return path) can go faster and a bit farther. (A transciever using all four pair of a Cat-5e, as of last year, could get gigabit ethernet across 30 meters.)

Frequency-domain techniques (like ADSL) can do still better. And coding schemes have been developed that get within 50% (turbo codes) or even 90%+ of the Shannon limit bit rate.

But what IS the shannon limit bit rate: It depends on a LOT of things. The biggest are:
- Length of the wire.
- Thickness of the wire.
- Quality of the dilectric around the wire.
- Interference coupled into the wire (i.e. how many other wires are in that bundle, what signals they're carrying, {for twisted pair} how tight the twists are and how they vary from conductor to conductor), how hot the wire is, etc.

You should be able to get gigabit rates to a box on your block with copper pair, with a small router there and fiber to the rest of the net. (This is "fiber to the curb".) For 10G or beyond you'll probably need CO-AX (ala cable TV) or fiber from the curb box as well - otherwise the curb boxes would need to be so close together that they get too costly - and you might as well have strung fiber from the one-per-neighborhood boxes.

(Maybe they'll push it a little farther. But I wouldn't hold my breath. Remeber that, in the US at least, you've typically got Cat-3 to the "curb" box which serves no more than 100 homes. If you're going to spring the bux dig it up and string 5e or 6 you might as well string some fiber. Later that can easily be upgraded to Tbits and beyond by transciever changes at the ends.)

Cool but...

[DR SoB]

"SolarFlare's chip will be used as evidence that 10G-bit over copper can be done, in anticipation of a draft IEEE standard to be developed later this year." "

Copper breaks down to easy, picks up to much interference, and is no good maintaining the speed over longer distances. They should concentrate on new technology instead of constantly trying to upgrade the old, now matter how much work you put into a '68 Mustang, it's always going to weigh a ton...

Re:Cool but...

[mystery_bowler]

Yes, but for a telecom, re-wiring is a pretty heavy investment. Depending on what state they are operating in there are different requirements for using unionized labor, there's literally tons of mechanical equipment involved, etc.

I'm not sure where the point of diminishing returns is, but it's still quite important that someone concentrate on taking the utmost advantage of copper since a lot of people are going to be stuck with it for a while.

Re:Cool but...

[FreeLinux]

Copper breaks down to easy, picks up to much interference, and is no good maintaining the speed over longer distances. They should concentrate on new technology instead of constantly trying to upgrade the old

It's funny but, that's what people said when networking vendors:

Increased modem speeds each time from 300bps to 56Kbps.
Introduced xDSL and then increased its speed.
Moved Token-Ring from 4Mbps to 16Mbps and then 100Mbps.
Move ethernet from 10Mbps to 100 Mbps to 1Gbps.

EETimes article with more technical details

[pm]

The EETimes carried this same story with more technical details and a few criticisms as a cover story in the week's paper edition. It's also available online here at the EEtimes website.

Re:Sign me up!

[ivan256]

That's funny, I'm quite pleased with the SLA I have with Worldcom, and quite turned off by the lack of SLA with any non telco options.

There's a difference between DSL and shitty DSL. Pick a company that *guarantees* the quality.

Now, if this stuff they're planning involves any encapsulation like PPPoE, or any "value added" services beyond a gateway and a block of static IP addresses, they can keep it, but I'd much prefer the phone company over the cable company any day otherwise. It's a lesser of two evils thing. When the phone company sells you something, you get what they sold you. Cable companies have a habit of changing the service you signed up for on a whim, and regularly. That combined the willingness to take responsibilty for problems (provided you pay for the right agreements) makes the phone company a no-brainer choice between those two options.

This keeps getting rehashed.

[FreeLinux]

10Gbps over copper was done, over limited distances, by Nortel three years ago. It's not new. In fact they are working with 40Gbps now, though not over copper, yet.

The technology ofr literally blistering speed is already available and hass been for some time. Additionally, it is not that expessive, relatively speaking, to offer speed that are significantly higher than todays broadband offerings. But, people keep bringing up the fibre to the home story and this is where the whole thing falls apart.

While new developments may indeed get fibre to the home but, no provider is going to "rewire". If they already have copper in the ground they are not going to upgrade. Why? Because of the cost.

Providers are already getting top dollar providing anything from 128Kbps (sometimes less) to 2Mbps. There is no incentive for them to make the massive capital outlay needed to bury fibre on routes that are already served by copper. It is unlikely that their customers will pay $100 per month versus the $50 that the providers already get for broadband so, there is no real demand to motivate the providers. Even new services like video on demand work adequately well over copper to negate the need for revamping the infrastructure.

No, providers will continue to offer the same services over their copper infrastructure and when things become saturated they will start to penalize people that use it the most. This is already happening with Comcast and AT&T.

Shut up about the last mile!

[Wesley Felter]

This article is about Cat5 cable. The last mile does not use Cat5 cable, so this article has nothing to do with getting a faster connection into your house. Let's mod all the "gee, I can download pr0n faster" comments as offtopic and get on with the real discussion about whether our processors are fast enough to drive 10Gbps.