When there's a killing in an Agatha Christie book, the sleuth looks for the Method, the Motive and the Opportunity. In networking, no-one makes a killing until we have the Product, the Need and the Price. A technology simply won't take off till the products are right, they meet a real need, and people can afford them.
Does 10 Gigabit Ethernet meet these criteria in the enterprise LAN? Vendors are starting to offer products, but is the offering complete enough? Is the price right? and can network managers be persuaded that they really want 10Gig and not just Gigabit Ethernet?
So far, 10Gig is available in core switches and fixed configuration boxes for the edge, but it still costs more than most users want to pay - especially as they can't yet see much need for it. Gigabit links are not yet filled up, so why rush?
"It's still a very early market, and a very small market, but it is growing strongly," said Seamus Crehan, director of Ethernet switch research at Dell’Oro Group. He says 9000 10Gig ports were shipped in the last quarter - an infinitessimal fraction the 43 million Ethernet ports shipped, but the high price of 10Gig means it now makes up two percent of the overall Ethernet market.
Slowly does itDespite the Ethernet tag, 10Gig is different from previous Ethernet revolutions. The price of a 10Gig port fell by 80 percent during 2003, which Crehan points out is a more aggressive fall than Fast Ethernet or Gigabit Ethernet managed - but for one reason only: the price was higher to start with.
"Fast Ethernet and Gigabit Ethernet were able to leverage existing technology," he says (adopting the FDDI and Fibre Channel physical layers), "so they came out at a low cost in the beginning."
Despite rapid reductions, 10Gig is still some way off the threshold where it can take off in the LAN: "History has shown us that, for volume adoption, one port at the new speed must be cost less than three times the cost of one port of the lower speed," says Crehan. With 10Gig ports averaging around $5000, we are well short of that, meaning that aggregating 1Gig links is still an attractive option.
Starting to do businessSo the 10Gig product announcements from Hewlett-Packard and others are really getting ready for a market that hasn't emerged yet.
So far, 10Gig has been pitched at service providers, and mostly placed in big core switches - but the hardware vendors want you to have it in your office. This is the key to big volumes, says Crehan: "The enterprise market tends to be elastic," he says, in contrast to the service providers who - at least since the dot-com boom - have tended to only buy what they need.
Outside the service provider market, the first users are large enterprises who are not price sensitive, such as CERN and oil exploration companies running large data centres and using grid computing, according to Peter Hulleman, senior research analyst at IDC.
After the large modular chassis switches which service providers use, and some enterprises have in their core, the next stage to get 10Gig into the enterprise LAN is to put uplinks on fixed configuration edge switches - vendors have done that, including Foundry, HP, Extreme, Cisco, Enterasys and HP. It's comparatively easy to do, and there's a sort-of justification for it.
Some vendors are pushing gigabit to the desktop heavily. Crehan says that more business PCs shipped with Gigabit than Fast Ethernet this year. So surely 10Gig is needed for their uplinks? Well, not necessarily, not if the Gigabit links aren't filling up, which - by and large - they aren't.
In any case, edge switches are more likely to link to a mid-tier level of switches, not directly to the core, and here adding 10Gig turns out to be trickier: the product fabrics for modular switches in this layer tend to be older, and not fast enough to support 10Gig at wire speed.
For example, Extreme has 10Gig in its Summit stackables and Black Diamond high end switches, but can't put it in its mid-range Alpines until they have a new fabric, predicted for next year. HP, similarly, is building a new mid-range fabric with technology from Riverstone.
Foundry may be the most aggressive here, with a forthcoming fixed-configuration aggregation switch that has eight 10Gig ports. "Foundry is very aggressive in the 10Gig space," says Crehan. "In general, they target the higher end of the market more."
But the compantition is very different from the days when Foundry was helping start the Gigabit Ethernet market - and one big difference is the number of start-ups. There were many Gigabit Ethernet start-ups including Foundry, Extreme, Alteon and other forgotten ones including Xlnt, Prominet and Acacia. In Ten Gig, there is only one: Force Ten.
"There is only one Ten Gigabit startup because the expected uptake of Ten Gig is much slower than Gigabit. That is realistic," says IDC's Hulleman. "Force Ten's balance sheet looks healthy, and it is focussing on the biggest and fastest products. Their most recent product has 572 Gig ports and 58 Ten Gig ports. That is quite impressive, from a technology perspective."
Could copper get it started? The arrival of copper connections jump-started Gigabit Ethernet, but the same isn't likely to happen to 10Gig, as it will never travel far on copper cables. Copper links are already available using Infiniband CX4 connectors, but that is "just an interim till 10GbaseT comes along," says Crehan.
Even then, 10Gig on copper will only be for stacking ports and connections within a datacentre or wiring closet. 10Gig backhaul and uplinks will have to be over fibre, and there the problem is adapting the standard to the large installed base of low-grade multimode fibre in buildings, particularly in the US. The LX4 standard offers a way to carry 10Gig over this fibre and now has an industry body, the LX4-TG to promote it.
"Gigabit backhauling was a real sweet spot," says Crehan. "There was big growth because there were plenty of people with FDDI grade fibre." At this stage, it's more complicated to do it with 10Gig.
Take your timeAs is so often the case, vendors are in a hurry and users should take their time. There is no point putting in 10Gig unless you have a likely need for it - and have the infrastructure to support it now. Otherwise, it's still too expensive.