By now, it should come as no surprise that networking is changing. The number of devices connected to the Internet has exploded. According to the recently released Cisco Visual Networking Index, by 2018 devices will generate more network traffic than PCs for the first time in history. Our devices have become mobile and our data centers have become hyperscale.
The arrival of mobility and hyperscale data centers has led to a need for much more flexibility and network programmability than ever before. While a classic wired network connected computers that mostly didn’t move, servers and devices in the modern network are always on the move. Our mobile devices, be they laptops, tablets or phones are constantly moving between various WiFi access points and broadband wireless data networks such as LTE. Yet we always expect our data to find us wherever and whenever we are connected. Because we are always moving the network must be constantly updated to keep track of where we are and how to connect to our devices. In the data center, servers are constantly being added and removed and workloads are constantly being balanced across hundreds of thousands of machines. This leads to constant changes to network configurations and the demand for standard approaches to monitor and program networks regardless of the equipment vendor.
Companies like Google and Facebook in building their hyperscale data centers have pushed to change networking in the same ways that they have changed computing. Rather than buying equipment from established suppliers they have opted to build their own gear by leveraging merchant silicon processors and ASICs, open source software and Asian manufacturing partners. Initially their servers were developed this way and more recently they have moved to developing their own storage and networking gear using the same model.
While Google, Facebook and other cloud scale providers are developing networking gear for their own internal consumption, a side effect of their efforts is both to demonstrate the feasibility of building networking equipment from merchant silicon chips and open source software, but also to enable other companies to follow suit. These change in the networking technology landscape makes networking equipment increasingly commoditized, and opens up the market to new levels of competition as new approaches, such as white box switching arrive and force a transition from a single-vendor to multi-vendor horizontally organized ecosystem where increased competition brings lower costs and more innovation.
Without the flexibility to program and configure the network – as well as reprogram and reconfigure – today’s networks would not be able to deliver the capabilities that we have come to expect.
Software-defined networks, typically referred to as SDN, have emerged as a possible answer, but not the only answer. The SDN approach is the simplify the forwarding hardware and decouple the control plane and, in most scenarios, run it as software services on standard servers. By decoupling the forwarding hardware and the control plane software, the network gains the flexibility to adapt to its own constantly changing state and provide for more frequent reconfiguration as well as the provisioning of new capabilities and services. Another possible approach is through white box switching or bare metal switching.
You’re probably familiar with the concept of the white box. After all, your PC is essentially a white box computer. It’s a piece of hardware that can run a variety of different operating systems and types of software. A white box switch is very similar. It’s a piece of hardware, purchased separately from the software. According to JP Morgan, this decoupling can save up to 52 percent over traditional, closed network switches.
In addition to the clear economic benefits, white box switches bring an inherent programmability to the network, making it more flexible and adaptable to the changing environment. This enables the network to more easily evolve over time, updating the switch software extends the life of the switch hardware and provides a mechanism for adding new features as they are needed in order to support new networking applications. The concept of white box switching also works to disaggregate the networking solution value chain allowing switch hardware vendors and switch software vendors to separately compete and innovate in hardware and software solutions. The network is no longer under the thumb of a single vendor, architecture or solution. With white box switching and SDN, the network might use a controller from one vendor, a set of white box switches from another vendor and run software offered by a third vendor. The network is no longer restricted by legacy solutions or cross-vendor compatibility.
Transitioning to white box switches can also serve as a means to support an orderly transition to SDN, with software updates to the white box switch supporting a classical switching environment initially, transitioning to a hybrid environment in the future then to an SDN environment as the technology matures and the organization is prepared to adopt it. Conceptually this is similar to reusing existing servers for new applications by reprovisioning them through loading new software on them.
While the transition to this new, decoupled, programmable, flexible and unrestricted network will not happen overnight, network operators are already starting to realize the benefits of white box switching. According to Dell’Oro Group, sales of white box switches outpaced the Ethernet switch market as a whole in 2013 and are on pace to do so again in 2014. As operators begin to explore this brave new networking world, they will discover that white box switching also enables them to benefit not only from lower costs and a path the SDN but also from the ability to increasingly integrate higher-level services into their programmable network switch solution, something we will explore in part 2 next week.
This article was originally published on ECN.
