Cisco's Chambers: Rival White Box Model 'Fatally Flawed'
In a wide-ranging interview, the Cisco CEO acknowledges an industry battle ahead but insists his company's architectural approach will prevail.
John Chambers loves to talk about "market transitions" and Cisco's uncanny ability to anticipate and prosper from them.
Cisco was founded to lead the market transition to multiprotocol routing in the late 1980s. In the 1990s it correctly called the transition from shared to switched local area networks while placing its big bet on Ethernet to the desktop (rather than ATM), and then it bet on IP as it moved into voice and video communications later in the decade. Now it sees the datacenter worlds of networking, storage, and servers coming together under a single architecture and strategic vision it calls Application-Centric Infrastructure, the centerpiece of a slick product, partner, and customer event at New York's Waldorf-Astoria hotel on November 6.
Cisco has a lot on the line with this market transition. By some accounts, what stands to happen in Cisco's core router and switch markets is similar to what has already happened (to some extent, at least) in storage and server hardware: commoditization. Network virtualization and software-defined networking (SDN) make it easier for customers to deploy cheaper white box alternatives to Cisco's high-end hardware, as the networking intelligence moves to software overlays. Google already builds its own switches using commodity hardware.
(Cisco's share price fell by more than 10 percent on November 13, after the company reported first-quarter financial results below analysts' expectations. Cisco also cut its revenue and earnings forecasts for the current quarter. Chambers blamed a "hard to read" economic environment and the chilling effects of the US government shutdown, among other factors, for the softness.)
Meantime, the Facebook-led Open Compute Project, whose members include Goldman Sachs, Fidelity Investments, and other companies with huge, highly sophisticated networks, is looking to extend that model to the masses, promising to share designs for low-cost network hardware that third-party manufacturers can bring to market. Vendors such as Intel, Mellanox, and Broadcom have announced open switch specifications via the Open Compute Project.
Cisco's November 6 New York City product announcement was intended to show it taking a leadership position in SDN while protecting its high-margin systems amid all the talk of commoditization. Cisco announced an SDN platform, developed under a company "spin-in" subsidiary called Insieme Networks, consisting of two main products.
The Application Policy Infrastructure Controller (APIC), running on Cisco's UCS servers, will create profiles for all the resources an application calls on, including bandwidth, storage, QoS, compute and load balancing, and then configure the network to provide the required resources. Cisco also introduced two switches under the new Nexus 9000 line: The Nexus 9508 is a 10/40 Gigabit Ethernet, 13-rack-unit chassis for end of row or aggregation; and a pair of Nexus 9300 switches, offering a mix of 1-, 10- and 40-GbE ports, are for top of rack. As my colleague Andrew Conry Murray noted in his coverage of the Cisco announcement, the APIC works with both a physical underlay and a network overlay. Currently, the underlay requires a Nexus 9000 switch as the spine node in a leaf/spine architecture.
Chambers: "This is the biggest move we've made and will have the most industry impact."
Cisco's and Chambers' message to customers: Bet even bigger on our integrated networking-server-storage architecture and we'll help you cut your operating costs (more on that later) and "future-proof" your IT systems.
"There will be a battle here" In an interview with InformationWeek following that announcement, the Cisco CEO adamantly dismissed the white box, commoditization threat and insisted -- as he did in an interview with InformationWeek in September -- that Cisco will win tech's next elimination round. He said:
There will be a battle here, which is fine, in terms of people pushing the white label approach or people who will push an SDN overlay. That [model] to me is fatally flawed. That's one we should have knocked off earlier. We have 25,000 software engineers. If this could be done in software we would have already done it. When you overlay networks, it becomes dramatically more expensive.
There are huge operational costs involved, no future-proofing whatsoever, and it's a problem for reliability and information security. It's impossible. What do [customers] want out of SDN? Programmability, virtualization, and the ease of bringing applications.
So what happened in the storage and server hardware markets won't happen in routing and switching? I asked. He responded:
The technologies you outline, storage, and servers, those technologies were dramatically underutilized. They were not integrated tightly into the network or tightly into applications or differentiated. And who gets the huge premium for providing network storage? Cisco-EMC, Cisco-NetApp. And who's the only one who gets a premium in the server market? Cisco UCS.
Customers look at total cost of ownership... What customers are after is time-to-applications and future-proofing their environments, whereas the majority of the cost is combining vendors that weren't designed to work together. And the only thing worse than vendors that weren't designed to work together are white boxes that weren't designed to work together.
Cisco has been making that case for some time, a case that Gartner called into question in a November 2010 report which argued that multivendor networks are less complex and ultimately less expensive to run than all-Cisco networks. Cisco continues to maintain that the Gartner report focuses too much on acquisition costs and not enough on long-term TCO.
Chambers continued on that theme:
The cost of operations in the average enterprise account is 60 to 80 percent, compared to capex, which is 20 to 40, probably 20 to 25. So your big savings is in the opex… Ask the operational person who has to tie together all of these individual white label boxes that were not only not designed to work together, but you have to be the one who troubleshoots across them, you have to be the one who understands the architecture, you have to be the one who knows the lifespan of those products will be only a couple of years -- not like the average Cisco product, which is five to seven years… This is giving your problem to someone else at a lower cost, and then five years later your IT organization gets the problem dumped back on them. And now they have to go clean up the mess.
We welcome your comments on this topic on our social media channels, or [contact us directly] with questions about the site.