As software-defined networking (SDN) and network functions virtualization (NFV) mature, what role will traditional network hardware play in cloud providers' data centers? How will their network hardware requirements change?
As these developments progress, there will continue to be a need for the four major categories of data center networking hardware we are accustomed to using: edge, core, top-of-rack and appliance devices. Each category will experience a distinct evolution to fulfill its role in the more virtualized data center model. In general, the amount of service processing intelligence required in hardware will be greater at the network's edges, both WAN- and server-facing, and the need for simplicity, economy, performance and scale will dominate in the core. The amount of control plane processing required in every type of node will decrease relative to today, as topology management and service controls migrate northbound to abstracted control plane systems.
For more information on SDN and NFV
Understanding the relationship between SDN and NFV
How SDN and NFV simplify network service chain provisioning
NFV: Another road to scalable networks
That said, there are still critical functions each type of node will continue to perform. At the WAN edge, routers will handle an important mix of virtual private network (VPN), tunneling and traffic engineering functions to support services like MPLS VPNs and Internet border routing. Inside the data center, critical functions include support for overlay network tunneling protocols, such as MPLS over generic routing encapsulation (GRE) and VXLAN, in addition to standard Layer 3 and Layer 2 protocols.
Core requirements in the data center will drive providers toward deploying networks with fewer switching tiers -- potentially collapsing from three tiers to two over time -- as well as higher bandwidths (100 Gigabit Ethernet [GbE] and up) and a much simpler and more scalable implementation of Layer 2 and Layer 3 services to better emulate a fabric topology.
Bandwidth demands will also steadily increase at the server edge from 10 GbE to 40 GbE, and the intelligence to support overlay network tunnels and gateways will be a requirement. Both the WAN- and server-facing nodes need the intelligence to interface with and transit between distinctly different domains -- such as WANs and data center interiors, or overlay virtual systems and data center interiors -- and require a different kind of hardware functionality than the fast, efficient and scalable core.
As the need to support secure and accelerated services from virtual systems expands, a greater number of virtualized appliance implementations will appear. Cloud providers will continue to need conventional appliances, but those devices will operate in more constrained roles and represent a smaller percentage of the total appliance mix.
Do you need advice? Submit your cloud computing questions for our cloud experts to answer.
This was first published in December 2013