Cisco Drops Hints About Insieme & SDN
It's an application-minded data-center fabric, one that's built for new cloud-networking trends and the imminent rise of software-defined networking (SDN). And of course it's got an Insieme-developed ASIC at its heart.
Separately, Cisco announced a less radical data-center fabric, based on a technology called Dynamic Fabric Automation (DFA), that's meant to be more of an upgrade to today's fabrics. It's a more conservative approach but one that Cisco says aggressively flattens and simplifies the data center. Some customers will favor that approach, so DFA shouldn't be thought of as a place-holder for Insieme, Cisco officials said.
(Side note: In baseball, "DFA" means "designated for assignment" -- a status assigned to a player that a team intends to get rid of within 10 days.)
The Insieme fabric will be available later this year. What's it made of? We don't know yet; Cisco only talked "vision" in Wednesday's press conference and didn't specify any Insieme products. However, the Nexus 7700, a big honkin' 100Gbit/s switch that also got announced Wednesday, will be usable inside the Insieme architecture, Cisco officals said.
The Insieme vision
Cisco revealed Insieme, or the idea of it, during its Cisco Live! customer event, being held in Orlando, Fla., this year. The press got an early preview at 8:30 a.m. Eastern time that featured Soni Jiandani, senior vice president of Insieme.
Insieme presents what Cisco calls an Application-Centric Infrastructure, one where the data center responds automatically to the needs of applications, doing things like moving virtual machines out of congested areas and migrating the necessary security and policy enforcement with them.
That's actually a goal of most SDN architectures. Cisco emphasized its use of an Insieme-developed ASIC, however, constrasting it with architectures relying on merchant chips from the likes of Broadcom Corp. -- a dig at Arista Networks Inc. and also at most other SDN architectures.
Jiandani described merchant-chip boxes as being hampered by the pace of the chips, which she said take three years between major generational shifts. The next generation of any ASIC also takes years, but Cisco's implication is that it's always ahead of the merchant-chip curve.
The hardware is important because network virtualization that's done in software will add complexity rather than reduce it, Jiandani claimed.
Insieme's ASIC purportedly will allow any type of encapsulation for moving virtual machines, for instance. There are multiple, competing protocols for doing this -- VXLAN from VMware Inc. and NVGRE from Microsoft Corp., for instance -- and Insieme's chip will be able to apply combinations of them at will.
Some implementations require external gateways to do something like that. Jiandani disparagingly referred to this as the "Christmas-tree baggage" required by merchant chips.
Insieme is also going to use some merchant chips, but only for basic functions, "rear-view mirror" types of tasks, as Jiandani put it. Jiandani also stressed that Insieme takes a view of the network as a whole -- what Cisco is calling a "system" view -- rather than working box-by-box. "There is nothing that is bringing a whole system together in these [other] models," she said.
Designated data-center fabric
Insieme specifics weren't disclosed. Cisco did go into more detail about the DFA, its more conventional fabric.
It's a revamp of the concepts in today's data-center fabrics, including the Juniper Networks Inc. QFabric. David Yen, the executive who once touted QFabric's superiority, is now a senior vice president at Cisco and was the one who introduced DFA in Wednesday's press conference.
The popular way to build a data center is in a leaf/spine architecture -- the leaves being Layer 2 switches at the top of each equipment rack, connected through a Layer 3 spine that's essentially the data-center core. Cisco is bringing the Layer 3 part down to the leaf nodes, creating a distributed gateway across the physical network edge.
DFA also connects every leaf node to every spine switch, so that no two virtual machines are more than two network hops apart. That adds predictability to so-called east/west traffic -- the communication between virtual machines that's causing the need for so many data-center changes in the first place.
DFA is due to be available in the fourth quarter of 2013.
Cisco did announce one specific product: the Nexus 7700 data-center switch, due to ship in July (i.e., in a couple of weeks).
It improves on the Nexus 7000 in raw numbers. The 7700 will have line cards that carry ten 100Gbit/s ports apiece, well more than the four ports that most other switches and routers are just starting to achieve. The 7000 is getting an upgrade, too, with six-port 100Gbit/s cards coming soom.
Arista has bragged about getting to 12 100Gbit/s ports per card, using on-board optics. Cisco didn't specify how it's getting to 10 ports. (See Arista's On-Board Optics Boost 100G Density.)
All told, the 18-slot version of the 7700, called the Nexus 7718, will be able to support 192 100Gbit/s ports or 384 40Gbit/s ports.
— Craig Matsumoto, Managing Editor, Light Reading