The move to 10Gig Ethernet is gathering speed, but it hasn't yet reached escape velocity, and for those involved with it, that can make for a frustrating business. To get an insider's view of how far it has got - and of what might come next - Techworld recently caught up with 10Gig pioneer Force10 Networks.
The signs of growth are clear, says Steve Garrison, Force10's corporate marketing VP: "There's 25 times as many Gigabit ports selling as 10Gig now. It was more like 50:1 two years ago. There are 5000 10Gig ports out there and the installed base is growing 20 percent a quarter - we are selling hundreds every quarter."
However, while big sales to leading edge users can bring headlines, as did CERN's decision to use 10Gig from core to WAN, there is the painful wait for the technology to take off in the mass market, which is where the real money can be made.
Garrison cites research by Dell-Oro on how the total system revenue for each successive Ethernet generation has built up in the past. The curves show a remarkable similarity, with each taking around three years to reach $1 billion, then the curve showing a 'knee' as over the next three years the technology goes mainstream and sales accelerate.
However, 10Gig's take-off was delayed both by the recession and its initial high cost, and Garrison admits that while it should still pass the billion mark relatively soon, it is still expensive, at maybe six to 10 times the per-port cost of Gig.
"Stackable 10Gig is around $3250 per port list price, chassis is $7000 to $8000 average selling price - that's an eight-fold reduction in three years," he says. "By comparison, line-rate Gig is $500 to $600 per port. Next year the general price decline should bring that knee in the 10Gig curve, and once it's only four times more it will take off.
"It's frustrating for all of us, that painful wait after launch, that three to six year wait for something to go mainstream."
Could copper be a boost?
Some have suggested that the emerging spec for 10Gig over 100m of copper UTP (namely 10GBase-T - there is already a short-range copper spec for clustering and the like, called 10GBase-CX4) could help accelerate things, but Garrison is sceptical.
"10Gig Base-T won't cause a big uptick, just more interest," he says. "Mostly people have fibre already, then you just have to make sure you have the right lasers."
He adds that there are already promising signs of 10Gig moving beyond the backbone, anyway. "The leading-edge data centres are moving to a 10Gig core and 10Gig to the storage," he says. "The highest end ones, where time is equivalent to revenue, are starting to do 10Gig to the servers. Plus you can drive 10Gig Ethernet 80km over dark fibre, so you don't need a Fibernet ring in a city like London, just dark fibre.
"We're now seeing people staging up to 10Gig aggregation, but what do you then have in the core - trunk 10Gigs together? So then people ask about scalability, and they are starting to ask about 100Gig Ethernet - that's maybe a 2009/10 release from IEEE. Most Internet exchanges are using Gig internally, and they are keen to go to 100Gig too."
Some Force10 users are already trunking 10Gig links, so in effect they already have 40Gig - a Japanese TV company is using trunked 10Gig for moving HDTV files, for example. Garrison says that the only snag is that there is still some uncertainty over exactly what will come next.
"There's two camps in the IEEE. The service provider camp wants 40Gig because of SDH, but the components people designed 40Gig stuff and nobody bought it. So a lot of the market want to go to 100 and make it simple. There's no standard yet for either, though - and there's also a group that wants to do 160Gig with a drop-down to 40."
He adds though that the standards process is now much less technology-driven than it has been in the past. "Standards bodies now look at the market opportunities before starting a new project, so 100Gig has to go 10km or it won't ship, then it's a cost and power trade-off," he explains.