Hyperscale companies like Facebook and Google, with their billions of users, are part of our daily lives. Behind the scenes, these companies have built dozens of massive data center networks to meet the needs of their customers. In doing so, they have changed the rules of the game to keep pace with the demands for data.
According to Gartner, “Hyperscale cloud providers (e.g., Facebook, Google, Netflix and Amazon) and others have massive scale data center networks and provide self-service offerings. In essence, the hyperscale web providers were forced to fill the operational innovation void that established networking vendors created. This required them to fundamentally re-examine how to operate networks that forged out-of-box thinking that has delivered radical contrast to traditional enterprise network operations."
The good news is that enterprises of any size can learn a thing or two from hyperscalers like Facebook and Google about data center networking. Aspects of their model can—and should—be replicated by the rest of us.
Nowadays, all modern businesses demand infrastructure that is agile, scalable and easy to manage in order to keep pace with new and changing requirements being placed on the business. The fact is, legacy data center networks are failing to meet these requirements as they remain highly static and require extensive manual intervention and operational overhead. It's no wonder businesses everywhere are increasingly adopting hyperscale-inspired network design principles to reduce complexity, capital expenditure (CapEx) and operational expenditure (OpEx).
What is hyperscale networking? Hyperscale networking is a design philosophy, of which attributes include a significant reduction in CapEx, a quantum leap in automation, and operational simplification. Hyperscalers adopted disaggregated networking models, which enabled them to outpace the vendor community in innovation, and also set a precedent: next-gen networking.
Each hyperscale data center is unique, but the design principles are applicable to a broader audience. All sorts of organisations, including software-as-a-service companies, telcos, government agencies, universities, large financial institutions and just about any business with a data center, can adopt hyperscale-inspired innovations to create a network that never goes down, can scale at the speed of demand as needed.
Design principle #1: Open networking switch hardware The first pillar of designing a hyperscale-inspired network is to procure ‘bare metal' switch hardware. Networking software, hardware and ASICs (application-specific integrated circuits) have traditionally been provided by vertically integrated vendors resulting in high prices and an extremely slow pace of innovation. This has finally started to change, with ‘merchant silicon' companies leading the effort towards a disaggregation model, and merchant silicon-based Ethernet switches have been steadily gaining market share over proprietary solutions.
Operating systems and software applications are procured separately, leading to tremendous innovation across the network stack, including in CPU, hardware, operating system and applications.
Much like ‘no frills' supermarket brands, bare metal switches are produced by the same manufacturers that build the hardware for legacy, incumbent switch vendors. Calculations estimate that using bare metal switches can reduce CapEx by up to 50 per cent. This is a significant reduction compared to using traditional networking vendor custom hardware, while the disaggregation of networking hardware and software also enables vendor choice, flexibility and increases competition.
Design principle #2: The software-defined approach to networking The second principle of hyperscale-inspired network design is to incorporate a software-defined network (SDN)-based architecture, which centralises network provisioning, troubleshooting and control functions, thus introducing new levels of network automation while simplifying network management.
SDN refers to the separation of the network control plane from the forwarding plane, enabling the network control to become directly programmable. SDN software moves network intelligence to a centralised controller for management and control, hence driving down management costs and network downtime.
According to Gartner*, “The bottom line is that we anticipate that enterprises following webscale networking principles will improve their device: admin ratios by more than 50%, and deliver services to the business at least 50% faster. Furthermore, we believe organisations that automate 70% of their network configuration changes will reduce the number of unplanned network outages by more than half, compared with those that automate less than 30%.
Design principle #3: Scale-out pod design Finally, instead of using a traditional core-aggregation-edge design, hyperscale operators take what's known as a ‘core and pod' approach. This consists of a routed core connected to a large number of independently designed pods. These pods contain the network, as well as the compute and storage. The advantage of a pod design is that they can be independently deployed and easily upgraded without affecting the core function of the network.
As the hyperscale philosophy is based on individual pod units, getting started for many organisations simply means choosing an appropriate project that is large enough to identify its own infrastructure build-out, for example, infrastructure-as-a-service, VDI or big data builds.
Until recently, hyperscale networking was out of reach for all but a few big players. However, companies like Big Switch Networks are enabling a much broader audience to adopt hyperscale-inspired networking. From my experience, the organisations that are finding the most success are embracing the design philosophy holistically and making judicious choices about the projects they can get to work on. Now is the time for hyperscale-inspired innovation at any scale!
By Mario Vecchio, managing director, Asia Pacific, Big Switch Networks