Most of the network processor and switch fabric startups of the 1999 era got nuked, so how is it that Dune Networks is still standing?

To hear executives tell it, the secret was simple: Dune actually managed to sell some chips.

"Almost any announcement of large enterprise vendors or large carrier vendors that are talking about large numbers and competing with, say, Cisco Systems Inc. (Nasdaq: CSCO) or Juniper Networks Inc. (NYSE: JNPR) -- you can assume they're using Dune devices," says Ori Aruj, vice president of marketing. That includes at least one seller of multichassis core routers, a category where Dune says it's seeing increased demand.

In 1999, startups were flocking into the market for network processors and switch fabrics, chips that would power the guts of routers and switches. But the dominoes fell: The CLECs collapsed, starving the switch/router startups of orders, and that in turn erased the market prospects for network processors. (See Switch Fabrics Storm the Market and Cogni-Gone?)

You hear a lot about network processor survivors Bay Microsystems Inc. , EZchip Technologies Ltd. (Nasdaq: EZCH), and Xelerated Inc. , but on the switch fabric side, the startups mostly died out, with just Dune left standing and independent.

Its competition in high-end switch fabrics -- the kind for chassis-based gear, not stackable Ethernet boxes -- remains limited, too, primarily to Applied Micro Circuits Corp. (Nasdaq: AMCC) and Broadcom Corp. (Nasdaq: BRCM).

Dune did more than just survive. It's recently profitable, and half the $45 million the company has raised in funding is still in the bank, Aruj says. (See Dune Digs Up $24M and Dune Networks Bags $12M.)

But what got the business going in the first place? It might have a lot to do with an early partnership with Marvell Technology Group Ltd. (Nasdaq: MRVL).

"They jointly developed the fabric that Marvell's MX devices and FX devices interface to," says Jag Bolaria, an analyst with The Linley Group . "So, as a result, they were able to go into some of the Marvell accounts and sell the fabric into them -- probably Extreme Networks Inc. (Nasdaq: EXTR), people like that."

Switch and router vendors don't like to talk about their use of outsiders' chips, so Dune doesn't have many specific jobs to point to. The company won't confirm that it's supplying chips to Extreme, or any other vendor.

So, how can you tell someone is using Dune? The product announcements earlier this week might provide some clues.

For instance, Dune brought out the FE600 fabric chip, which has four times the capacity of its predecessor but can fit in the same designs. So, if a router vendor suddenly announces a quadrupling of switching capacity in a system, without the need to change any linecards -- there's a chance it's because of Dune, Aruj says.

Dune's other product announcements this week likewise ratcheted up the speed. The P220 and P230 -- fabric access chips that go on linecards to communicate with the FE600, which sits on a separate switching card -- handle 80 Gbit/s in one direction, up from 40 Gbit/s for previous devices. (See Dune Scales Switch Chips.)

Dune expects to increase that capacity to 100 Gbit/s, to target 100-Gbit/s Ethernet cards once the IEEE standards are set. But the company seems to have had motivation to get devices to market more quickly than that.

"We have seen data centers where people are using 80-Gbit/s ports -- it's sort of a proprietary standard -- to connect boxes at high speeds," Aruj says.

The key for Dune, and the reason it's managed to stick around, is that the switch fabric is central to a system's design. A change of network processors would probably force a company to design new linecards; a change of switch fabric more likely means redesigning the whole box.

"The switch fabric market is not a huge market, but it is strategic," Aruj says. "Once you're in there, it's almost like you're part of the company."

Bolaria thinks Dune is likely to continue going it alone, rather than getting acquired. For one thing, he doesn't believe Dune has any wins at Cisco, Juniper, or Alcatel-Lucent (NYSE: ALU) -- which limits its value to other chipmakers. And Dune probably wouldn't sell itself to any systems vendor, since that would destroy its business with other OEMs.

There's that partnership with Marvell to consider, but Bolaria doesn't expect that to lead to a merger. "I believe Dune probably thinks they can get better valuation for themselves than getting acquired by Marvell," he says. "But these aren't the days you talk about an IPO."

— Craig Matsumoto, West Coast Editor, Light Reading