To bring a DevOps approach to the network, one expert suggests enterprises take a page from the cloud-provider playbook. Here's how data center pods can support a NetOps evolution.
Networking's NetOps movement aims to bring the benefits of DevOps -- automation, collaboration, programmability and virtualization -- to the enterprise network, with the goal of building infrastructure that readily adapts to today's ever-shifting business needs. Proponents of an industry-wide NetOps transition say network managers must take a multi-pronged approach, making changes across tools, technology, people and processes.
In the data center, Gartner analyst Andrew Lerner -- an early advocate of NetOps, or, as he and his colleagues at Gartner refer to it, NetOps 2.0 -- suggests enterprises emulate the DevOps-friendly designs of hyperscale organizations. Instead of designing and managing a data center network as one massive entity -- building it out all at once with periodic, network-wide upgrades -- he recommends incremental rollouts of smaller data center pods.
A pod, or zone, is a largely self-contained unit with as few as two network switches, Lerner said -- ideally with just 100 to 400 ports total. A central backbone connects the data center pods, performing only basic transport functions.
A pod, or zone, is a largely self-contained unit with as few as two network switches.
This type of design enjoys widespread usage among web-scale cloud providers, which have found it "the only way to sustain high growth rates and high availability," Lerner said.
The modular design makes the data center network more responsive to innovation, allowing managers to add capacity on demand -- simply deploying additional pods to shoulder new workloads as necessary. It also lets them design fit-for-purpose network zones, with pod-specific parameters.
"Maybe you need to support high-frequency trading or self-service for a specific set of workloads but not others," Lerner said. "Or maybe you want some to have different change requirements than the rest."
Data center pods also enable staggered upgrades -- limiting hardware and software changes to one confined area at a time, thereby containing the blast radius of any outages. Theoretically, overall resiliency and availability improve, as well, as a configuration bug in one pod shouldn't affect its neighbors.
Importantly, the pod architecture also supports the core DevOps tenet of continuous processes. With self-contained zones, network pros can more easily embrace iterative development, continuous testing and blue/green deployment in the network. Rather than waiting several years for a major refresh, data center pods constantly improve, individually and as a group. "Each generation of pod can copy successful characteristics of prior pods," Lerner said.
