How Automation Changes Network Management

What it is

Network automation means using software to do network tasks that a person would otherwise do by hand.

For many years, engineers managed devices one at a time. They opened a command line, typed commands, saved the result and moved on. That works for a handful of devices and breaks down with hundreds or thousands.

A script or tool connects to devices, makes changes, collects information and checks the result. Programmability is the ability of a device to be controlled this way, through interfaces designed for software instead of for a person at a keyboard.

How it works

Automation replaces repeated manual steps with a tested process built on APIs, data formats and version control.

What automation improves

Speed. A task such as adding a VLAN to fifty switches shrinks from hours to minutes.

Consistency. Automation applies the same standard configuration every time. This reduces configuration drift, which is the slow divergence of devices from their intended settings as people make small manual changes over the years.

Fewer human errors. Many outages come from a typing mistake or a forgotten step. A tested script does not forget.

Better compliance and checking. Tools can compare each device's real configuration with the approved one and report any difference. They can run the same checks every day, not once a year.

Faster troubleshooting and visibility. Automation can collect show output from many devices at once and put it side by side. Modern devices can also stream telemetry, which pushes data continuously to a collector instead of waiting to be asked.

That gives much finer detail than the traditional approach of polling devices every few minutes with SNMP.

The building blocks

Automation depends on a few technologies.

• APIs: programmable interfaces that let software control a device or controller. REST APIs, covered in a later topic, are the most common.
• Data formats: structured text that software can read, such as JSON, XML and YAML.
• Data models: a formal description of the configuration and status data that a device holds. YANG is the language used to write them.
• Protocols for model-driven management: NETCONF and RESTCONF are used to read and change data described by YANG models.
• Configuration management tools: Ansible and Terraform, covered in their own topic, apply desired configurations to many systems.
• Version control: a system such as Git stores the scripts and templates and records who changed what and when, so a bad change can be undone.
• Controllers: central systems that hold the intended state of the network and push it to devices.

A useful idea is infrastructure as code. The configuration is written as text files, kept under version control and applied by a tool.

A key property: idempotence

A good automation task is idempotent, which means running it once or ten times leaves the system in the same state.

If a VLAN already exists, the task sees that and changes nothing. This makes it safe to run again, and it makes the task describe the desired result rather than a list of steps.

What automation does not do

Automation is a powerful tool, but it has limits.

• It can spread a mistake to every device in seconds. A bad template is just as fast as a good one.
• It needs testing, review and a way to roll back.
• It requires new skills, such as reading data formats and using version control.
• It does not replace understanding. Engineers must still know what the network should do, so they can design and verify it.

The role of the engineer therefore changes. Less time goes on typing repetitive commands, and more goes on design, on writing and reviewing automation, and on checking the results.