It's been a hard slog to 10Gig Ethernet. Even now, problems remain, chief among them, power consumption and the cost and physical size of today's short-reach XENPAK and X2 connection modules, which restricts how many ports a switch can have.
On top of that, the various attempts to get the physical format right have caused confusion to manufacturers and network buyers alike. For example, with five different 10Gig modules on the market and three fibre types to choose from, that's a boggling fifteen ways that you can connect a network device.
In addition, even the latest XFP module only allows 32 ports in a standard rack chassis, and saves little on power and cost. Meanwhile, Gigabit Ethernet is up to 48 ports per box and sales are flying ahead.
So chip companies are looking for the Holy Grail that will drive 10Gig uptake. On the way there, they are developing 10Gig devices that enable more and more functionality to be moved out of the laser module and onto the linecard in the switch where it is cheaper, and to take out as much complexity as possible.
Yet more new formats are emerging too, chief among them a 10Gig version of the SFP Gigabit fibre module called SFP+, which is being evaluated by the IEEE for standardisation as SFF-8431. The industry is hoping that this time, it will have got the physical format right at last.
There are big problems here, though. The signal frequency needed to send and receive data at 10Gig is vulnerable to all sorts of interference, both between the module and the linecard, and within the chips themselves. Until now, this has usually been addressed by adding silicon within the module, such as a clock and data recovery (CDR) chip to correct timing errors.
Chip-makers push hardware
Several of the companies promoting SFP+ and the use of less expensive silicon in the module are actually chip designers, not switch or optical manufacturers. That's not as unlikely as it sounds, says Mitch Kahn, marketing VP at Canadian chip developer, Quake Technologies (which was recently acquired by AMCC). He points out that the chip developers can't sell their 10Gig chips in volume unless the roadblocks holding back 10Gig adoption in the enterprise can be dismantled.
"As a company that's completely focused on 10Gig, it is important for us to see the market grow, so we are creating technology that enables these new switches to be built," he argues.
That technology is a new chip, the QT2035S, which Quake claims is the first to integrate a standard XAUI-interface 10Gig Ethernet PHY (or physical interface), with SFP+ signal processing support.
The chip includes programmable electronic dispersion compensation (EDC) electronics capable of equalising the connection between the PCB and the optics. In effect, it analyses the 10Gig signal and calculates how much anti-distortion to add to the signal to counteract the real distortion so the laser receives a clean signal ready for transmission.
Kahn says that EDC is the same technology that enables 10Gig-LRM to run 220m over multimode fibre, by compensating for the relatively poor quality of the optical fibre.
In this case, as well as compensating for the fibre, it also enables cheaper optical components to be used, such as those designed for 2Gbit and 4Gbit Fibre Channel, plus it allows the optical module to be much smaller.
"The advantages of SFP+ really come down to cost and density," Kahn says. "You've kicked out all the electronics you can and moved them to the PCB. For example, the clock recovery chip is $50, consumes 0.5W and takes up space, and can be eliminated.
"Gigabit has already migrated to SFP, it's also used for 2, 4, 8 and 10Gig Fibre Channel - half the SFP market is Fibre Channel and we anticipate the same for SFP+. "
Kahn claims that, where an XFP module currently wholesales at $350 (and retails for many times that), an SFP+ module should wholesale for under $100 and eventually around $50.
A challenge to copper
He adds that in his view, 10Gig is set to be the first generation of Ethernet where copper cabling won't dominate.
"We're not developing 10GBase-T," he says. "I don't know whether I'd call it a competitor, or an enabler for more fibre. It is so hard to do that i think you'll definitely see a different mix of copper and fibre in this generation - it was 85 percent copper in Gig, but it will be 50 percent at 10Gig."
Part of the problem is power consumption, which requires cooling and restricts port density.
"The best we're hearing for copper is 4W for 35m range, or 6 to 10W for 100m," he says. "XFP is 2.5W, X2 is 4W but that's not a good comparison because it has extra chips.
"There's a latency issue with copper too - I would never bet against copper though!"
The growing interest in 10Gig over copper also reflects the fact that it's not just WANs and MANs that need the extra speed now.
"The 10Gig market has shifted towards datacentre applications, we expect it to be 50 percent datacentre next year with the rest divided between intermediate range and metro," Kahn says.
"The CDR chip was the biggest item to remove - I'm not sure there's much else to go, as then you're down to about nothing in the module. Instead the innovation will move to the lasers, with 1310 and 1550nm lasers moving to the less expensive VCSEL process currently used for 850nm lasers."
It's going to take time for SFP+ switches to appear, though - the Quake chip goes into sample production this month, but Kahn acknowledges that the first switches won't arrive until 2007
"Within two years we will see new gear coming out with SFP+, but there will still be lots of installed X2," he adds.