Lecture Notes: AI and Machine Learning in Network Operations

Introduction

• Application of AI and machine learning to prevent IT infrastructure failures.
• Focus on eliminating noise and reducing remediation time.
• OCTA's role in transforming network operations.

Overview of OCTA

• Founded by the speaker, who is the CEO.
• Applies AI/ML to operational data in networks, servers, and infrastructure.
• Operates in data centers, service provider networks, and large-scale LLM infrastructures.

Challenges in Network Operations

• Historically, network operations have been noisy and reactive.
• Presence of vast data across various layers (cloud, data center, 5G, etc.).
• Operators have been using siloed tools for data mining.

OCTA's AI and ML Solutions

• Uses real-time algorithms, largely unsupervised, to find insights from data.
• Detects anomalies, predicts issues, and correlates events to reduce noise.
• Proven to decrease tickets by 70-90% and detection time from 47 minutes to 1 minute.
• Platform is software-only, scalable, and can be deployed on-premise or as a SaaS.

Integration and Data Collection

• Collects data from infrastructure components (switches, routers, servers, etc.).
• Integrates with data lakes like Prometheus and Splunk.
• Designed for big data and Telemetry streaming.

Unique Algorithms and Approach

• Built custom algorithms after finding open-source solutions inadequate.
• Focuses on misbehavior detection across TCP, optical, HTTP layers.
• Automates actions such as ticket generation and issue remediation.

Use Cases and Applications

• Observability and AI Ops in hybrid cloud, data center, 5G environments, etc.
• Use cases include:
  • TCP retransmissions and congestion correlation.
  • Optical misbehavior detection ahead of failures.
  • Post-change verification for BGP and other changes.
  • LLM infrastructure job metrics analysis.
  • Synthetic probing for quick issue identification.

AI Ops and the Role of LLMs

• AI Ops is a reality, already implemented at large scales.
• LLMs play a supportive role, but not central in OCTA’s strategy.
• Unsupervised ML is used for log analysis and anomaly detection.

Market Impact and Future

• Growing adoption and interest in AI Ops, significant ROI reported by users.
• Gartner projects a major increase in enterprise adoption by 2030.
• OCTA aims to help enterprises mature their network operations with AI Ops.

Conclusion

• AI Ops in the network is mature and beneficial.
• OCTA offers a comprehensive approach to transform network operations.
• Encouragement to engage with OCTA for operational transformation.