360° Intelligent IT Operations: From Scripts to AI‑Driven Automation

360° Intelligent IT Operations: From Scripts to AI‑Driven Automation

Speaker Sun Jie, a veteran operations expert with more than a decade of experience in system, cloud, and data‑center management, shares insights from his work in foreign firms, internet companies, e‑commerce, and large enterprises.

He emphasizes that AI in operations is currently an assistive tool, similar to early autonomous driving, and that most intelligent operations are still in the exploration stage.

1. Building a Comprehensive and Scientific IT Operations Management System

Insufficient overall recognition of the IT department; many leaders still view it as a cost center rather than a profit‑generating unit.

Heavy workload for operations staff; without automation, a small team cannot investigate beyond firefighting.

Lack of holistic situational awareness; thousands of metrics exist but no unified view for intelligent perception and trend prediction.

Inadequate ability to adjust services and resources based on business needs, leading to slow fault‑resolution processes.

Desired operational goals include comprehensive performance management, unified resource management, timely fault alerts, and integrated visualisation.

Real‑time performance monitoring with dynamic thresholds.

Unified resource view across cloud‑based services for rational allocation.

Accurate, low‑latency alarm notifications.

Consolidated dashboards that aggregate data from multiple monitoring subsystems.

Core concerns:

Cross‑region, multi‑data‑center platform for unified IT management.

Fine‑grained monitoring and centralized visualisation to improve efficiency.

Proactive problem prevention and rapid root‑cause tracing to reduce labor costs.

Multi‑dimensional reports to support decision‑making and trend forecasting.

Integration of business‑service perspective, platform extensibility, and big‑data analytics.

Protection and optimisation of IT assets through cloud‑based consolidation.

2. Panorama Business Service Management

In the era of digital transformation, IT services must respond quickly to support business.

Monitoring granularity must be fine enough to capture anomalies on a single curve.

Separate business‑oriented management from user‑oriented management, with clear permission boundaries.

Comprehensive and extensible data collection is essential for scientific decision‑making.

Building a business‑driven monitoring platform enables unified display, management, and scheduling across the entire request‑response chain.

Linking resource monitoring to business impact allows rapid identification of affected services and users when performance degrades.

3. Big‑Data‑Based Log Analysis and Multi‑Dimensional Reporting

Using a big‑data platform for log collection, aggregation, and correlation enables accurate fault localisation, performance optimisation, and predictive alerts.

Collected network, data‑center, server, cloud, and video‑surveillance data are stored in a performance‑management database (PMDB) and modelled for resource‑KPI analysis.

Data acquisition can be passive or active; preprocessing tags relevant metrics and formats noisy logs.

Performance thresholds may be static or dynamic, calculated from historical peaks to guide resource allocation.

Event diagnosis relies on time‑series correlation to filter relevant logs and pinpoint root causes.

Algorithmic assistance, even simple open‑source models, can accelerate intelligent operations.

Data aggregation compresses and standardises collected information; ingestion can use full‑load HDFS and incremental Kafka streams.

Multi‑dimensional reports generated daily, weekly, or monthly feed management with clear visualisations of incidents, impact, and resource planning.

Key concerns include performance analysis, capacity planning, and automated configuration, e.g., forecasting storage needs based on business growth.

4. Evolution of IT Monitoring Platforms

Three generations: network‑centric monitoring (1990s), infrastructure‑centric monitoring (hosts, storage, OS, middleware), and application‑centric monitoring that focuses on user experience and high‑availability with real‑time diagnostics.

5. Fault Management and Autonomous Healing

Before automation, alarm volume caused anxiety; the goal now is to filter and distil essential alerts.

Target attributes: simplicity (timely response, automated analysis), intelligence (machine‑learning‑driven fault classification), and depth (knowledge‑base sharing for faster resolution).

Machine learning requires large datasets for training; human‑in‑the‑loop labelling improves model accuracy.

Building a strategy knowledge base enables less‑experienced engineers to handle incidents efficiently.

Ultimately, intelligent operations aim to reduce human dependence, allowing machines to make autonomous judgments, decisions, and actions.