If, like most people, you don't yet have more than a few strands of Gigabit Ethernet about the place, you might wonder why so much effort is going into even higher speeds. And it's not just 10Gig either - router and switch developers are now looking to 40Gig and 100Gig, debating which will come next.
"40-100 is a good debate," says Chandra Kopparapu, VP and general manager of the service provider and multilayer switching business unit at Foundry Networks.
"The technology exists to do 40Gigabyte products now, but 100 is not yet at a commercial level, it would need some sort of wavelength muxing so there's more work to do. We think 40Gigabyte Ethernet may be purchased for some applications, for example on an OC-78 compatible interface it would give you a SONET physical interface running Ethernet."
SONET connectivity is already driving some of the uptake for 10Gig Ethernet, says Marc Randall, president and CEO of 10Gig specialist Force10 Networks. "In metro applications, we are really seeing demand for SONET start to slow down and demand for Ethernet interfaces increasing," he adds.
"There's a little bit of Ethernet over SONET but not a lot. Ethernet over fibre is definitely the trend, and we will see a move among service providers to Ethernet over MPLS. The world is really coming down to Ethernet and SONET. In the core of the network, FDDI is gone and Token Ring is gone."
Dell'Oro analyst Seamus Crehan points out that so far, Ethernet generations have come at 30 to 36 month intervals, with each one growing to become a $1 billion market - the accepted mark of it breaking out into the mainstream - over the same time frame.
On this analysis, 10Gig should hit $1bn by the end of 2005, then it will take off in volume in 2006-7, with 100Gig appearing at the end of 2005 or the start of 2006.
The exception was 10Gig, he says: its ramp-up was delayed by the recession and it was expensive, offering no cost advantage. Previous regenerations had offered 10 times the speed of what went before, at less than 10 times the cost, so were immediately worth using as trunks.
"Gigabit is mainstream in backbones, data centres and aggregation, but it's not at the desktop yet - that's only just starting," Crehan says. "We started to see 10Gig in 2002, given a similar evolution you're looking to 2008 for the next generation to go mainstream.
"Who knows if the switch market will go 40Gig or 100Gig? Either way, it's a long way out. If we do see 40Gig, we might also see 10Gig in metro areas, but metro networking is a broad term that's more popular in Asia than elsewhere, with Japan the biggest market for aggregating broadband."
Crehan points out that the vast majority of Ethernet port shipments, and therefore revenues, come from switches rather than routers, so the test of 10Gig's success will be switch deployments.
As for the 40/100 debate, he says: "We will see routers on 40Gig with OC-78 linecards, but for switches it's always been factors of 10. Who needs 100 though? That's an awful lot of bandwidth once you go above 10Gig."
The other factor holding up both Gig and 10Gig adoption has been the media, but Chandra Kopparapu says this is changing fast:
"10Gig is predominantly fibre now, you will see copper products around the end of this year, and 40 will be fibre-only," he says. "Gigabyte copper is predominant now - the inflection point was last year, with a big part of that being Gigabyte to servers."
"10Gig looks and feels like Ethernet, so people don't care what's inside or what the physical interface is, but it remains switch-to-switch so you have to look at the distance limitations - copper will have a 10-30m limit."
Kopparapu points out too that 10Gig is already worth having on price grounds: "10Gig was $50,000 a port two years ago, now it's $3500 so we have crossed the point where one 10Gig port costs the same as 10 one-Gig ports. "Gigabit Ethernet pricing is seeing a fairly steady decline, so enterprise-class switches might be $100 for 100Mbit, $400 for Gigabit.
"Some Gigabit copper products are close to 10/100 pricing but they may only give 10/100 performance thanks to the cost of non-blocking backplanes - it'll be another year for full Gigabit performance at the same price."
The reality though is that there are plenty of backbones running on 100Mbit Fast Ethernet, never mind Gigabit, says Peter Tarrant, who sees quite a few LANs in his role as VP marketing at NAS company ONStor.
"Gigabit is still coming in, and 10 Gig is still 12 to 18 months away - (a) it's need, and (b) it's price," he says. "When it comes it will be no big deal to us to upgrade our ports, but we're not seeing very much 10 Gig at all yet."
Even grid and cluster CPU interconnects are still Gigabit at best, admits Marc Randall, with some still using 10/100. He says that 10Gig is coming in for trunking between clusters though, and may act as an enabler for IP storage schemes such as iSCSI; however, he adds that IP storage is merely incremental so far, and is not replacing existing SANs.
Looking forward though, he believes that "10 Gigabit will be the enabler for new technologies, for example video-on-demand will take up more bits, and enterprises will need bigger pipes to let more customers in."