Just after VMworld, NoJitter editor and all around smart guy, Eric Krapf, sent me this link to a Wall St Journal article article that talks about the efforts of both Facebook and Google to build their own network infrastructure and obviate the need for traditional vendors such as Intel and Cisco. Eric may have his flaws as a Chicago sports fan, particularly given the dominance of all Boston teams over the past decade, but he recognizes interesting topics to blog about.
The theme of this article has been the source of much confusion and misunderstanding since SDNs became the biggest tech buzz word since "cloud". The thesis is that organizations are now able to deploy a commodity network layer and then create their own software layer for all of the functions they used to pay companies like Cisco, Brocade and Extreme for. Or, at the extreme (pardon the pun), some companies could just build their own switches from scratch. This trend would be the end of networking as we know it. Network infrastructure would be commoditized and the valuation of companies like Cisco would be lower than the combined win total of the Cubs and White Sox.
So is this true? Are we seeing the end of networking as we know it? Will all companies just build their own hardware or at the very least software? I do think there's truth to the story but I think it's misleading. The fact is, companies like Google, Facebook, Baidu and Amazon have always built much of their own hardware. Some have even dictated to the hardware manufacturers how to build products that are optimized for their environments.
These companies can do this because they have very sophisticated engineering teams that can build custom features that create competitive advantage. Google has hundreds of network engineers that do nothing but tinker with custom builds to create unique features. The same for the other big web giants. So I think the article is true in that those companies are trying to circumvent traditional hardware by buying off the shelf white boxes from the likes of Accton and Quanta--but they can afford to do so, and any small advantage they get can pay huge dividends.
However, I believe the appeal of this networking strategy is limited to a handful of the top-tier organizations that have both the resources and the need to create custom network infrastructure.
There's another tier of companies below the "custom hardware" line, and that's large enterprises that want the flexibility of the software model but do not have the resources internally to write their own operating system and need some support. For these organizations, a crop of startups have emerged that offer disruptive technologies. These vendors, such as Pica8, Cumulus, Plexxi, Pluribus and the most disruptive vendor, Arista, offer software running on off-the-shelf hardware. This provides customers with a feature rich, high quality operating system and technical support. For organizations that do not have "Google like" resources, the OS and support can provide some flexibility without the overhead of a totally custom environment. This model will likely appeal to large financial organizations, some service providers and universities. Large organizations, but not the largest of the large.
For the rest of the tens of thousands of companies out there, creating custom network infrastructure and lowering the cost of hardware isn't a top priority. Rather, keeping a network secure and up and running is the most important element. I know much of the media has focused on how the cost of traditional network hardware is too high and commoditization is coming, but that's something that's been predicted now for over a decade and it's never happened. Nor do I think it's needed.
Consider that 20% of the overall budget for running a data center is used for hardware, which includes, servers, storage and network infrastructure. The network accounts for about 15-20% of that spend, so a total of 3-5% of overall data center TCO. On the other hand, personnel costs account for an average of 40% of data center TCO. For organizations that don't need the custom features but require the resiliency, security and reliability of traditional equipment, they are willing to pay the premium to buy networking from not only Cisco but also Brocade, Juniper, Extreme and the like.
The Journal article that Eric pointed out is certainly an interesting read and creates great debate, but the appeal of custom network hardware will remain limited to those companies that it always has. For the rest of the world, staying with the tried and true is a safer, smarter choice--at least for now.
Follow Zeus Kerravala on Twitter and Google+!
@zkerravala
Zeus Kerravala on Google+
