Network Automation

AI application to networks

Accelerates service provisioning.

• Makes networks understand service intentions.

Reduces the probability of human errors.

• Automates complex networks.

Predicts and quickly rectifies faults.

• Enables networks to optimize themselves.

Improves efficiency of detecting threats.

• Enables networks to predict threats.

A self-driving network

A network that:

• Accepts guidelines from the network operator.

• Self-discovers the components.

• Organizes and configures itself.

• Monitors itself using probes and other techniques.

• Auto-detects and auto-enables new customers.

• Automatically monitors and updates service delivery.

• Diagnoses itself using machine learning and heals itself.

• Periodically reports by itself.

How can this become true?

• Telemetry.

• Multidimensional views.

• Automation.

• Declarative intent.

• Decision making.

Telemetry

• Obtain information from the network, the data, services, etc.

• Real-time vs. Statistical.

• Raw vs. pre-processed.

• Key Performance Indicators.

The Aggregation Scenario

Support the integration of different data flows.

• Open.

• Automated.

• Secure.

• Scalable.

Deal with heterogeneity at all levels.

• Data sources.

• Data consumers.

• Data models.

• Deployment styles.

• Supporting infrastructures.

Not just data

• Metadata becomes essential, including semantic mappings.

• What seems to claim for a data stream ontology.

• Not that far away: data modeling is a first step.

The Actuating (Control) Stream

OAM actions at a wide variety of different domains

• Challenging, given the current state-of-the-art.

Initial strategies.

• Domain-specific.

• Recommendation systems.

• Autonomic protocols.

SBA approaches and capability models.

• Reusable functionality description.

• Abstractions of network element functionalities usable as building blocks.

• Combined to provide more powerful features.

• Registration mechanisms to support CI/CD.

• Inter-domain collaboration for E2E management.

Network Automation

Declarative Intent

Decision Making

Rule-based learning

• If ‘X’ happens, do ‘Y’

• Pros

  • Straightforward programming.

  • Easy to predict and refine.

• Cons

  • Slow work.

  • At scale, hard to manage.

Machine learning

• Models, learning, percentage.

• Pros

  • Can become creative.

  • Fastest way to learn complex behavior.

• Cons

  • Can come to strange conclusions.

  • Hard to know what it knows.

A transformation of the skillset

Network engineering → Service design.

Network knowhow → Algorithm development.

The networks get out of the way.

• SLAs are automatically met.

Networks adapt, react, anticipate.

• Security becomes Good Guy ‘Bot vs Bad Guy‘ Bot.

Network automation

Remove human from the loop of usual network management tasks.

• Provisioning.

• Detection.

• Diagnosis.

• Remediation.

In reality, is not to fully remove the humans, but remove them from the low-level painful tasks.

Why?

• Humans are expensive.

• Reliable?

• How to scale?

Closing the Closed Loops

The use of closed loops is common everywhere.

• Automatics have been around for a long time.

• An essential aspect of industrial processes.

Not only about offering network data.

• An integral monitoring data substrate.

Well-defined data flow semantics.

• Data models for sources and consumers.

• Registry, discovery and dynamic orchestration.

• Full data sovereignty.

Going beyond.

• Key-Value Indicators distillation.

• Network-hosted AI and learning mechanisms.

• Support for serverless in-network computing.