Fujitsu Packs in 10GigE
Of course you wouldn't; Fujitsu's $200 claim does seem exaggerated. Still, with 10-Gbit/s Ethernet switches just starting to ship and with per-port prices in the five-figure range, Fujitsu's Layer 2 switch chip has the potential to scrunch down prices more quickly than expected, particularly in enterprise-related areas such as storage networking and blade servers.
The chip was first announced June 25 (see Fujitsu Intros 10-GigE Switch Chip) and is being demonstrated at this week's Networld+Interop Tokyo. Its starring feature is density: The chip packs a whopping 12 ports of 10-Gbit/s Ethernet into a device measuring 35mm on a side. By contrast, early systems from the likes of Cisco Systems Inc. (Nasdaq: CSCO) haven't put more than four ports of 10-Gbit/s Ethernet into one box. Awkwardly named the MB87Q3050 (something catchier is on the way, assuming it passes a trademark check), the chip has drawn more attention than expected, says Marwan Majid, Fujitsu senior marketing manager.
"It's quite surprising, because we've always been debating [within Fujitsu] when 10-Gbit/s Ethernet is going to happen," he says. Majid thinks interest in the chip signals that carriers are ready to start thinking about spending money, which has OEMs preparing next-generation systems.
A big question is whether the chip will rewrite the rules for the OEMs that have announced 10-Gbit/s systems -- namely, Cisco, Extreme Networks Inc. (Nasdaq: EXTR), Force10 Networks Inc., Foundry Networks Inc. (Nasdaq: FDRY), and Riverstone Networks Inc. (Nasdaq: RSTN).
Those vendors' per-port prices range from $6,300 to more than $50,000 (see Riverstone Fuels 10GigE Price War) -- which would make Fujitsu's $200 claim seem as plausible as a faster-than-light engine.
As it happens, Fujitsu isn't counting the optics necessary to build a box like those mentioned above. The optics for one long-haul Xenpak interface can run $5,000, according to a Foundry representative. So optics alone would bring Fujitsu's chip in at the low end of the aforementioned per-port claims.
Also, Fujitsu isn't necessarily chasing long-haul or Layer 3 applications. Majid sees the chip being used in areas such as aggregation servers or box-to-box interconnects; in other words, $200 per port arguably points to an inexpensive Layer 2 box. (Still, Majid stresses the chip has a future in IP storage, because Fujitsu has gotten the switching latency down to 450 nanoseconds.)
So, while the switch chip does sport QOS and virtual LAN capabilities, it's not as involved as an IP switch fabric. It's closer to the Ethernet switch chips offered by Broadcom Corp. (Nasdaq: BRCM) and Marvell Technology Group Ltd. (Nasdaq: MRVL) but on a more advanced level; those two companies haven't yet seen a reason to produce a device with that much 10-Gbit/s Ethernet on it (see Marvell Readies GigE Attack).
Still, the Fujitsu announcement isn't all bluster. The chip's density is noteworthy by itself and has earned Fujitsu a presenter's spot at Hot Chips, a deeply technical chip symposium held every August. "The integration is fairly impressive, in terms of the number of 10-Gigabit ports on a single chip," says Tony Hsu, product marketing manager for Marvell.
Further 10-Gbit/s announcements are likely from Fujitsu, as the company's chip-design team -- which works on ASICs for OEMs including Foundry -- apparently has quite a portfolio to show off. Majid hinted the company will soon announce silicon chips that run serial 10-Gbit/s signals. Most chips, including the MB87Q3050, split 10-Gbit/s signals into multiple slower lanes, because logic circuits can't handle 10 Gbit/s yet; serial 10-Gbit/s capability would put Fujitsu alongside Broadcom in the hunt to push the speed limits of silicon.
— Craig Matsumoto, Senior Editor, Light Reading