Making Gig to 10Gig a seamless step
Can cheaper technology for 10Gig fibre displace copper LANs in our hearts?
If your network is one of the many that has fibre installed, whether for a 100Mbit/s or 1Gbit/s LAN backbone or perhaps a SAN, you're probably wondering how much it is going to cost to replace it all with higher capacity fibre once your users start demanding even faster speeds.
And they will, says Dr Paul Voois, president and CEO of semiconductor company Clariphy: "As soon as you get to the typical Fortune 5000 deployment, every new PC is Gigabit-enabled, and new switches are usually Gig with 10Gig uplinks.
"Fibre is the only game in town for 10Gig today. If 10Gig to the desktop happens, that's likely to be a copper play, but that's well over five years away."
Clariphy's interest is all this is that it is developing what Voois claims will answer to the upgrader's dilemma - a PHY (physical interface) chip for fibre optic networking that will enable today's multimode fibre installations to be upgraded to 10Gig via the upcoming 10GBase-LRM standard.
Not only will 10Gig over LRM be cheaper than today's LX4 spec for 10Gig fibre, he argues, but it will go further than the 10GBase-T spec for 10Gig over copper and generate less heat too.
A power struggle
"10GBase-T will have its place, but it has a 55m reach limit on CAT6, and probably takes 8W to 10W of power per port, with several microseconds latency," Voois says. "Fibre can reach 500m easily on 3W per port. Then it comes to cost, and the cost of fibre is coming down rapidly.
"It's an extreme argument, but sooner or later the data rates will overwhelm copper. Maybe we're wrong and that's at 100Gig, but at some point it's true."
However, Clariphy's plans - it hopes to have its chip out in the third quarter of 2006, with actual networking products based on it following by year-end - highlight the somewhat confused future of 10Gig networking.
Voois reckons there's 100 million multimode fibre ports out there today with another 15 million added each year, and all of it installed for lower speed networks. Around two-thirds of fibre port shipments are Ethernet and one-third Fibre Channel, he says.
"Fibre is the only medium that works for the enterprise backbone," he continues. "In the data centre there is a mix of fibre and copper, but in large enterprises fibre is dominant."
Twisted pair is still fighting
However, not everyone agrees with all those forecasts. HP ProCurve's Mark Thomson says that over time, the power consumption of 10GBase-T should settle out below that of 10Gig on fibre, although he acknowledges that the physics of 10Gig will prevent it falling to 1Gig's 1W per port. And he adds that optics will always be a step up in cost from copper.
"What customers really want is 10Gig over twisted pair," he says. "With 10GBase-T the objective was 100m, but it's dependent on the quality of the cable so you might not get that.
"The DSP [digital signal processing] requirements just to make that work are scary. It wasn't even possible two or three years ago, but probably in a year from now we will have products you can have confidence in."
Those advances in DSP technology are also what makes 10GBase-LRM feasible, says Voois. They allow the PHY, or physical interface module, to recover more of the light signal from the fibre, and he claims they'll also allow less expensive optics to be used than LX4 would need.
Fee PHY fo fum
Voois says another important factor to watch out for is the PHY. Currently, most equipment manufacturers are using or moving to the compact X2 and XFP form factors, which allow them to increase port density over the older Xenpak type, but Voois says the one to watch for is the even more compact SFP+.
He says that SFP+ will cut cost and increase density still further because it uses a cheaper serial interface instead of parallel XAUI between module and line card, and because it moves all the decoding electronics out of the PHY module and onto the card.
There's a lot of "if" in there of course, and there are a lot of other companies, also working on how to make 10Gig run further and cheaper over both fibre and copper - notable PHY developers include Quake, Phyworks, Vitesse and Scintera.
But even if copper does catch up eventually, it does look as if fibre will indeed be the dominant 10Gig medium for a good while yet. Forecasts from market analyst Dell'Oro say that Gig copper port shipments only overtook Gig fibre ports in 2004, after three years of fibre having higher sales volumes, and there is little reason t think 10Gig will be any different.
And more importantly for those who already have Gig fibre, it should be reusable for 10Gig - although you probably won't get quite the same range.