Facebook, whose demand for servers and switches is matched only by the world's demand for cat pictures, has been busy creating reference designs for scaled networking resources. By opening these designs, or at least promising to -- as of late June the Wiki contained only tumbleweeds -- others may duplicate and benefit from them. What impact will this have on the networking landscape in the short and long term? Ultimately, it comes down to whether you are a consumer or a supplier of IT systems, but first it's worth understanding a little about the switch, and why it's important.
Facebook's Wedge, while powerful, is a relatively straightforward top-of-rack design that wouldn't draw attention in any data center. The special part? It's not what has been put into the box, but what has been removed.
Facebook Wedge designed for flexibility
To that end, the hardware comprises a basic top-of-rack switch with a merchant application-specific integrated chip (ASIC), 16 40 Gigabit Ethernet (GbE) ports, dual-power supplies and fans in a 1 rack-unit enclosure. The conventional processing power comes from the Open Compute Project's "Group Hug," a micro server board on a stick that is connected via a standard PCI Express bus.
There are no exotic interconnects, no black magic engineering, proprietary buses or closed software. It is the simplest way of moving data between legions of servers.
The component that is missing, however, is proprietary software. By creating FBOSS, its own software stack based upon open source Linux, Facebook can customize to meet its own needs. This enables the software to operate in lock step with the rest of its application and server suite. By throwing the platform to open source, Facebook may be able to reduce the ongoing development costs as others will contribute to the platform for the greater good.
Another key advantage to Facebook -- and any other enterprise that chooses to take on this approach -- is freedom from the shackles of the supply chain. The journey a commercial switch takes between its fabrication and the data center rack is a long and convoluted one. If a critical component supplier can't make its deadline, or, for example, a fire knocks out another vendor for six months, the impact these incidents would have on a switch vendor would be serious indeed. Required software revisions, regression testing and re-stocking of the distribution and hardware maintenance channel take time. The customer may then face unpalatable delays as the wheels of the large vendor turn.
Facebook freed from conventional supply chain
With its latest strategy, Facebook has freed itself from many of these restrictions. With direct control over the fundamental parts of the switch, an alternative component may be selected, tested and rolled into the next software update without impacting deployment schedules. Without any middlemen, the time to resolution is massively reduced.
Where does this leave the market as a whole? In the short term, there will be little impact, but other Internet-scale organizations will follow in the footsteps of Facebook and Google. Customers at this high end of the market will have fewer reasons to pay hardware vendors premium prices as the dual incentives of reduced cost and operational flexibility increase. Yet, adopting the "roll your own" approach to network infrastructure takes special consideration. At the moment, most enterprises still need the development resources of a medium-sized vendor or integrator at their disposal.
The hardware still has to be built and a software image delivered that is fit for the purpose.
Develop your own switch or have someone make one for you?
The point of demarcation comes when you determine whether to craft your own innovation or pay someone else to do it for you. If you lease a platform or services from someone else, then the Open Compute Project will be of little interest, except perhaps as an indicator of how forward- looking and competitive potential delivery partners are. For anyone delivering services, this method of purchasing and assembling network components to suit budgets and demands will make a lot of sense. Network vendors will lose traction in the bigger accounts, forcing them to focus on volume rather than high-end customers.
For those enterprise customers that continue to take an active part in the selection and delivery of IT network services, they will find renewed attention from vendors as the big fish customers swim off. This commoditization is, of course, pretty inevitable, but it will ultimately lead to lower operational costs for all, regardless of whether they choose to build their own network or not.