Why would Cisco be developing a new data center fabric product line based on merchant silicon when its Nexus switches and FabricPath software are already available?
Observers are still speculating about last week's news that Cisco is developing a new Nexus switch, and perhaps fabric switches, based on Broadcom silicon instead of internally developed ASICs. Cisco won't discuss the unannounced products, so an explanation from the company isn't available.
Some analysts believe Nexus/FabricPath is incomplete, that it doesn't take the data center to a completely flat, one-layer fabric. That's why Cisco needs to develop a new line, which sources say it is doing under the code name "Jawbreaker."
And doing so with merchant silicon reduces the time and cost of developing custom ASICs, and speeds time-to-market - something of vital importance now that Juniper has rolled out its QFabric line from its Stratus fabric project.
"FabricPath reduces complexity, but from three tiers to two," says Zeus Kerravala, an analyst at the Yankee Group. "Stratus goes down to one."
Paving the way for the flat network
Jawbreaker may also be an attempt by Cisco to freeze the market; it's not expected to ship until late 2012 but Juniper's QFabric ships later this year.
Cisco is also expected to soon announce the Nexus 3000, a low-latency, high-density 1-rack-unit 10 Gigabit Ethernet switch specifically designed for market trading. The Nexus 3000 is based on Broadcom's Trident chipset, sources said, while Jawbreaker is based on Broadcom's upcoming Trident+ devices.
Nexus 3000 is not part of the Jawbreaker fabric, the sources said. But Cisco's been losing business to Arista and Juniper in financial trading departments, so the Nexus 3000 is intended to strike back, analysts say.
But basing any product on merchant silicon is a "huge departure" for Cisco, Kerravala says.
"It signals that they are feeling some kind of competitive pressure," Kerravala says. "But ASICs give them a competitive advantage" due to their proprietary technology.
Kerravala believes follow-on versions of Jawbreaker and the Nexus 3000 may be fitted with ASICs instead of off-the-shelf processors. Other analysts believe Cisco's decision to use merchant silicon reflects a confidence in the performance of the devices, and frugality.
"Since every other switching vendor -- including 'performance-oriented' firms like Arista, Force10, and Juniper -- use merchant silicon, Cisco now feels like it is good enough as well," says Jon Oltsik, principal analyst at Enterprise Strategy Group. "I think we are also seeing a financial decision here. Cisco needs to lower prices but protect margins. The only way to do this is lower its costs, thus merchant silicon. I actually think this is a wise decision."
As for Jawbreaker in particular, Oltsik believes Nexus/FabricPath is optimized more for Fibre Channel-over-Ethernet (FCoE) penetration than for flattening a data center fabric. Cisco is an aggressive evangelist of FCoE for virtualizing storage, running it as another traffic type over Ethernet.
"The Nexus/FabricPath stuff... was originally designed for FCoE, not Data Center Bridging/TRILL-like functionality, so my hunch is that Jawbreaker is a next-generation version of FabricPath," Oltsik says. "In other words, I can't see Cisco offering both. In my view, the goal of Jawbreaker is a flat network with storage coming along for the ride -- very similar to QFabric in this sense."
Indeed, Juniper sees it that way too. In a comment on Oltsik's blog on the Network World Web site, Andy Ingram, vice president of product marketing in Juniper's fabric and switching group , says Cisco's FabricPath fails to adequately accommodate Layer 3 traffic in data centers while working around the shortcomings of scaling Ethernet in non-blocking data center networks.
QFabric, meanwhile, "changed the scaling model" of an Ethernet switch to distributed and federated elements rather than multiple Ethernet switches. By "rethinking the data plane, the control plane, and the management plane," QFabric scales to more than 6,000 ports and 44Tbps of cross-sectional bandwidth while maintaining the operational model of a single switch, Ingram says.
"It would appear that even Cisco agrees with Juniper's approach," Ingram states in his post. "Cisco's 'Jawbreaker' project is eschewing the FabricPath approach and is attempting to emulate the QFabric switch."
Even the current Nexus 5000 and 7000 switches are architecturally inconsistent, says another Cisco competitor. Jawbreaker is an attempt to correct that, while the Nexus 3000 is a time-to-market play to make up ground against Arista and Juniper in financial trading.
"The N7k and N5k were designed for different purposes -- the 7k was 'big fast modular box, feature rich, price point not much of an object, etc.' -- the 5k was 'get FCoE to market, support (Unified Computing Systems),'" the competitor says. "Getting these all to play together in a way that looks and feels like a single switch from a CLI perspective and from a forwarding plane perspective, given their divergent genealogy, is damn near impossible. Thus they do a new architecture to achieve time-to-market against Juniper as it's the only way they can cost-effectively deliver the capability."
Join the CIO Australia group on LinkedIn. The group is open to CIOs, IT Directors, COOs, CTOs and senior IT managers.