22 .

Network automation is a broad category of techniques and methods used to automate network-related tasks, ranging from simple scripts for routine tasks to more complex automation platforms.

For example, a Python script can be used to reliably perform configuration changes on large numbers of devices in a fraction of the time required for manual configuration.

Traditional network devices perform a variety of functions on top of forwarding messages, such as building routing/ARP/MAC address tables, using Syslog to log events, and using SSH to accept remote CLI connections.

The various functions can be divided into three logical planes: the Data Plane, the Control Plane, and the Management Plane.

The Data Plane includes all functions directly related to forwarding messages over the network: receiving a message on one interface, performing any necessary processing, and then forwarding it out of another interface.

The Control Plane controls the Data Plane. Functions in the Control Plane are not directly involved in the process of forwarding messages but instead perform necessary overhead work to enable the Data Plane’s operations.

The Management Plane includes a variety of functions that don’t directly influence the forwarding of messages—functions related to configuring, managing, and monitoring network devices.

Traditional network architectures use a distributed Control Plane—the “brains” of the network (the Control Plane) are distributed among each network device. For example, each router uses OSPF to learn routes and build a routing table.

SDN takes a different approach, centralizing some or all of the Control Plane functions in a controller. This is called a centralized Control Plane.

In SDN architecture, each network device’s role is simply to forward messages according to the controller’s instructions. Although the Control Plane is centralized, the Data Plane remains distributed among the network devices.

SDN facilitates the programmatic control of the network through applications that interact with the SDN controller, resulting in a three-layer architecture consisting of the Application, Control, and Infrastructure Layers.

The Control Layer translates high-level requirements from the Application Layer into actionable instructions for the network devices.

The Infrastructure Layer consists of network devices like routers and switches that execute the command received from the Control Layer.

Communication between the three layers is achieved using application programming interfaces (APIs) and various communication protocols.

The interface between the Application and Control Layers is the northbound interface (NBI). It typically uses a representational state transfer (REST) API with HTTP messages.

The interface between the Control and Infrastructure Layers is the southbound interface (SBI). A variety of APIs and communication protocols can be used in the SBI, such as OpenFlow, NETCONF, OpFlex, and traditional protocols like SSH and SNMP.

SDN isn’t a single solution. Cisco’s SDN solutions include SD-Access for wired and wireless campus LANs, SD-WAN for WAN networks, and Application Centric Infrastructure (ACI) for data center networks.

These SDN solutions work by building a virtual network of tunnels (the overlay) on top of the underlying physical network (the underlay). The combination of virtual and physical networks is called the fabric.

Software-Defined Access (SD-Access) is Cisco’s SDN solution for campus LANs. The SD-Access fabric consists of a physical underlay of switches and a virtual overlay of tunnels using Virtual Extensible LAN (VXLAN).

Cisco Catalyst Center, formerly called Digital Network Architecture (DNA) Center, functions as the SDN controller in SD-Access.

Software-Defined WAN (SD-WAN) is Cisco’s SDN solution for WANs. SD-WAN creates an overlay of IPsec tunnels over any physical WAN underlay: the internet, MPLS, cellular 4G/5G, satellite, etc.

Application-Centric Infrastructure (ACI) is Cisco’s data center SDN solution. Like SD-Access, ACI creates an overlay of VXLAN tunnels over the underlay, which is a physical spine-leaf network.

The SDN controller used in ACI is called the Application Policy Infrastructure Controller (APIC).

Artificial intelligence (AI) refers to the simulation of intelligence in computers, allowing them to analyze data, identify patterns, and make predictions or take actions based on those insights.

Machine learning (ML) is a field within AI that allows computers to learn on their own, without requiring explicit programming.

With ML, computers can learn from vast data sets in a few different ways:

• Supervised learning—The ML algorithm is trained on labeled data sets.

• Unsupervised learning—The ML algorithm is trained on unlabeled data sets.

• Reinforcement learning—The ML algorithm learns by interacting with an environment and receiving positive or negative feedback.

Deep learning (DL) is a subset of machine learning that uses artificial neural networks to analyze and learn from large amounts of data. These neural networks can extract more complex patterns and relationships from data than traditional ML algorithms.

Predictive and generative AI are two important applications of ML and DL.

Predictive AI uses historical data to predict future events, such as weather forecasts and stock market predictions.

Generative AI leverages ML and DL to create new content, such as text and image generation.

ML and DL can be used to analyze vast amounts of network data to uncover patterns, anomalies, and insights.

Predictive AI has applications in network operations such as traffic forecasting, predictive maintenance, capacity planning, and security threat prediction.