The networking world is changing in three fundamental ways, all of which threaten Cisco’s dominance and market share in the world of switches, routers, and other gear that send packets streaming around our broadband and data center networks. Cisco is responding to the threats with a multi-pronged strategy. Its most visible aspect is much-hyped startup Insieme, the $100 million spin-in that the company created to address the data center. But Cisco’s strategy doesn’t stop there. Let me lay it out for you.
Computing has changed in the past decade, which has led to Google-style massively scaled-out data centers that serve one application, be it a cloud platform or Google. A decade ago, a data center was measured in thousands or maybe tens of thousands of square feet, but today they are measured in hundreds of thousands of square feet and sited next to power plants. For example, Apple’s new data center will measure 500,000 square feet and have 184,000 square feet of raised floor dedicated to servers. These massive server farms mirror the growth of the Internet, and inside they contain a similar web of traffic flowing from machine to machine. It was a shift Cisco failed to see coming, and its gear didn’t make the jump into the server farms customers were buying.
As part of that change, those behind the data centers have looked at the cost of buying Cisco gear (or that made by Juniper)and said: “No, thanks.” Instead they built their own—as Google did using chips from Fulcrum (now bought by Intel) or Broadcom—or they turned to companies such as Arista Networks to provide lower-cost hardware using merchant silicon.
This reduced the cost of boxes but provided more overhead in the form of engineering talent to manage the boxes. Arista’s software made it a better buy for those less interested in managing the networking at the hardware level, and it was fast. Thus, the rise of merchant silicon and the need for the data center to scale rapidly have created problems for Cisco. The biggest blow may prove to be the rise of software-defined networking and the OpenFlow protocol.
Using OpenFlow or other controllers running proprietary software, data center operators can separate the physical network infrastructure from the management of that network. This makes it easy to program and manage networks without ever having to understand a proprietary lower-level programming language. The combination of all these trends, plus the rise of cheap gear using merchant silicon from Cumulus Networks or Pica8, makes it hard to see why the big data center operators would buy Cisco gear in the future.
While all these shifts are occurring in the networking world, it’s important for people to recognize that the networking world encompasses a wide spectrum of users. On one end, there is the data center, the fiber routes between data centers, the long-haul networks that connect to the consumer last mile, and cell phone towers, as well as the consumer home network. On the other side, there is the service-provider network that provides the connectivity via wired or wireless connections. They, too, have data centers, fiber routes, and complicated central offices, plus other edge-network components. And while aspects of the data centers and the carrier network worlds are converging, it’s going to take a long time before things become interchangeable—if they ever do.
Cisco has products in all these spaces, and all are threatened by the rise of software-defined networking. This is why in an interview, David Ward, chief technology officer of Cisco’s service provider group, tells me: “There is an SDN vision at Cisco that is done and defined. And Insieme will be part of that, but it will not be the complete response.” Insieme is part of the data center strategy, which is where the threat to Cisco is the most urgent, but it’s not the only strategy.
Ward joined Cisco from Juniper in January to help create a portfolio of products that mitigate the threat of software-defined networks. Judging from his comments to me on Wednesday and a presentation at the Open Networking Summit, the strategy involves a Cisco-style level of complexity that incorporates OpenFlow but also will require other protocols and standards delivered via modular software-based building blocks.
The vision will rely on Cisco’s doing two things correctly: One is convincing service providers that having a fatter, rather than flatter, networking infrastructure—with multiple management and orchestration layers—will serve their needs better. Second, the company must figure out how to deliver good software that’s aimed at higher-level developers.
When it comes to convincing service providers they need more, not less, on the networking side, Cisco has an advantage. Service providers are worried about the costs of running their networks, but they are running multiple networks (think 3G, Wi-Fi, and LTE) and want to use software-defined networking to make them more agile and to eliminate gear, as opposed to focusing solely on lowering the cost per bit. Google, on the other hand, ripped out its intra-data center network and replaced it with one that’s OpenFlow-based and is cheaper to build and run.
On the first issue, Cisco will be fine. It’s selling this vision of service providers using the data they already have to tell applications about the state of the network in ways that could improve their service. It may even end up being something carriers could charge for. But the approach also means that Cisco moves up the stack in the software game. Sure, it writes plenty of quality embedded-systems software, but it isn’t known for the quality of its higher-level programming efforts.
Cisco is buying expertise in that area. Ward points to the acquisitions of LineSider and of BNI Video to indicate how serious Cisco is about software-defined networks and embedding that into its product portfolio. Ward is fighting the battle to keep service providers on board with Cisco’s gear, aided by the carriers’ reluctance to buy into the data center’s commodity view of the world. Insieme has a tougher battle to fight, but so far, it has $100 million and some smart people.
Also from GigaOM:
The Future of Wi-Fi in the Enterprise (subscription required)