How long until we get 400 Gbps ethernet?

[Mark R]

Just an idle Saturday question of purely academic interest. I don't particularly need such equipment at the current time.

While 40G ethernet seems vaguely practical, albeit niche - current versions of 100G look ludicrously impractical, not least because of insane power consumption and sheer size of the electronic modules (meaning far less total bandwidth available per rack slot than 40 G). It's not particularly cheap either.

Still, in a couple of years, 2nd gen 100G modules should make things a lot more practical.

But what of beyond?

 

[cmetz]

It appears right now that we're approaching the limit to how many bits per second we can move in a single channel. We're approaching the limits of what electrical circuits can do, and it's getting more and more of a problem every time you need to convert between optical and electrical domains. Much like we're doing with CPUs, I think greater degrees of parallelism will have to become more common in the future of networking, and we're going to start having to deal with the problems that introduces.

Power consumption, size, and cost can all be brought down given the inevitable march of process technology. The bigger question is whether we're about to hit a wall with the physics of the problem.

 

[VirtualLarry]

It appears right now that we're approaching the limit to how many bits per second we can move in a single channel.

I thought optical technology had a long ways to go, and that the limitation was the transceivers, not the fiber.

 

[Mark R]

I thought optical technology had a long ways to go, and that the limitation was the transceivers, not the fiber.

That's pretty much true.

Fiber has about 8 THz of bandwidth. Current systems can squeeze 1.6 TBps onto a fiber, by coupling 160x 10 Gbps transceivers onto a single fiber; each transceiver has a different wavelength (separated from its neighbors by 400 picometers/50 GHz) and optical diffraction gratings are used as wavelength selective on/off ramps).

As can be seen 1.6 Tbps out of 8 THz bandwidth is inefficient. There's a lot of interest in the use of advanced modulation techniques, such as phase-shift modulation, or combined polarisation/phase shift modulation. Initial research suggests it should be possible to get 100 Gbps onto a single 50 GHz channel - potentially boosting the capacity of a single fiber to 16 Tbps. However, the complexity of such proposed transceivers is staggering.

 

[cmetz]

VirtualLarry, like I said, it's the electrical side. Not just the transcievers, but building electrical circuits to do nontrivial things at those data rates becomes infeasible too (at least as far as we know). The fiber can handle it, but with no way to get bits on and off the fiber we have a problem, at least for a single channel. WDM many channels in parallel and we can scale.

 

[mammador]

this may seem strange, but surely high end Ethernet is only needed for specific applications/uses, right? A corporate WAN of, say, IBM or GE may use an ATM or 10 Gigabit bandwidth, but does a home user need 40 Gigabit bandwidth?

 

[Geofram]

this may seem strange, but surely high end Ethernet is only needed for specific applications/uses, right? A corporate WAN of, say, IBM or GE may use an ATM or 10 Gigabit bandwidth, but does a home user need 40 Gigabit bandwidth?

Need is a very relative term.

But really, this is computers we're talking about. As long as something isn't happening instantly, someone will be trying to find a way to make it do it faster.

I don't "need" to have my 30GB movies transfer between machines in a few seconds, rather than minutes, but it would be nice.

Of course, in the home sector, it will require faster hard drives to be standard before a faster network will do much.

 

[spidey07]

this may seem strange, but surely high end Ethernet is only needed for specific applications/uses, right? A corporate WAN of, say, IBM or GE may use an ATM or 10 Gigabit bandwidth, but does a home user need 40 Gigabit bandwidth?

10 gig is exploding in the data center, as in folks can't put enough of it in fast enough. The same will happen with 40 gig once the price comes down.

 

[mammador]

Need is a very relative term.

But really, this is computers we're talking about. As long as something isn't happening instantly, someone will be trying to find a way to make it do it faster.

I don't "need" to have my 30GB movies transfer between machines in a few seconds, rather than minutes, but it would be nice.

Of course, in the home sector, it will require faster hard drives to be standard before a faster network will do much.

maybe, but computers are always built to satisfy needs anyhow. And these needs depend often on the mode of use. Given how most home users utilise computer networks, I don't see how 40 Gigabit would meet their applications needs.

 

[Crusty]

maybe, but computers are always built to satisfy needs anyhow. And these needs depend often on the mode of use. Given how most home users utilise computer networks, I don't see how 40 Gigabit would meet their applications needs.

I want to backup my multi-TB file server remotely and I don't want to wait a week for it to finish.

 

[mammador]

I want to backup my multi-TB file server remotely and I don't want to wait a week for it to finish.

maybe, but i can't see ISPs providing high-end Ethernet when most home computer users aren't into networking and frankly don't need that degree of bandwidth. it's basic supply and demand, why provide something that has a limited demand? Note that most computer users globally only have light Internet and wider networking needs, why have high-end Ethernet for only Facebook or Twitter?

 

[spidey07]

maybe, but i can't see ISPs providing high-end Ethernet when most home computer users aren't into networking and frankly don't need that degree of bandwidth. it's basic supply and demand, why provide something that has a limited demand? Note that most computer users globally only have light Internet and wider networking needs, why have high-end Ethernet for only Facebook or Twitter?

You're not seeing what is driving all the demand for such speeds. Business, always has been, always will. I already mentioned that business can't install 10gig ports fast enough to keep up with demand from the applications that require them. Mainly high concentrations of VMs (think thousands).

The consumer/home market is an after thought to networking.

 

[mammador]

well that was my basic and initial point. it will largely be corporate LANs/WANs that will get this new standard, as the demand for it is higher. IT companies are profit-driven like any other, and of course the greater demand and hence profit is in the business sector.

 

[cmetz]

+1 to spidey07 - gigabit to your servers is so last decade. It's now gigabit to the desktop and 10Gb/s to your servers. You really need a 100Gb/s switch interconnect to make the math work out on that, but 40Gb/s is what it looks like we're going to have for a while.

10Gb/s pricing is coming down on a nice curve now, it won't be too much longer before it's feasible for prosumers.

 

[spidey07]

+1 to spidey07 - gigabit to your servers is so last decade. It's now gigabit to the desktop and 10Gb/s to your servers. You really need a 100Gb/s switch interconnect to make the math work out on that, but 40Gb/s is what it looks like we're going to have for a while.

10Gb/s pricing is coming down on a nice curve now, it won't be too much longer before it's feasible for prosumers.

We have seen this trend since the first 10base-t switch was out. The demand is nothing new.

20k for a 12 port 10/100 switch? Fucking bargain that was.