How AI Enables Unattended Cloud Server Management and Self‑Service Automation

In the cloud‑native era, enterprises face growing complexity in IT operations, diverse business demands, and massive operational data, creating urgent needs for precise alerts, intelligent anomaly diagnosis, root‑cause analysis, prediction, and automated remediation.

On September 26, 2020, Alibaba senior technical expert Teng Shengbo presented at the GOPS Global Operations Conference in Shenzhen, sharing how Alibaba Cloud’s Elastic Compute team uses AI to achieve unattended, self‑service management of cloud servers, simplifying instance management and ensuring stable, efficient service.

Why do cloud servers need unattended management?

Alibaba Cloud’s unattended self‑service practice

The AI and data behind unattended operations

1. Why do cloud servers need unattended management?

Operations encompass both infrastructure software services and human services, targeting business teams that use the infrastructure. Cloud IaaS now serves developers and ops teams. With over a million users running workloads on Alibaba Cloud, the platform faces three common pain points in ECS instance operations: high communication cost to understand issues, long manual resolution times, and lack of transparency in customer actions.

To avoid linear growth in support costs as user scale expands, Alibaba Cloud applies AI to empower automated operations, aiming for unattended cloud servers similar to trends in unmanned retail and driving.

Leveraging more than a decade of ECS operational experience and machine‑learning‑driven behavior analysis, Alibaba Cloud built an unattended architecture offering self‑diagnosis, self‑repair, self‑optimization, and self‑operation, reducing management complexity and ensuring stable, efficient instance services.

2. Unattended Self‑Service Practice

Cloud IaaS operations can be divided into platform‑side and customer‑side tasks. Platform‑side tasks (invisible to users) involve data center, physical equipment, resource virtualization, scheduling, and live migration. Customer‑side tasks (visible to users) include instance modifications, scaling, monitoring, ticket handling, and orchestration.

Our unattended architecture provides a suite of self‑service capabilities. Broadly, Alibaba Cloud’s self‑service covers ECS instances, lifecycle management, system management & automation, and marketplace/ecosystem, as illustrated below.

Specifically, the self‑service offers intelligent diagnosis, automated repair, optimization recommendations, best‑practice templates, and event automation, covering about 80% of common ECS issues and reducing average resolution time from hours to minutes without human intervention or privacy risk.

ECS Intelligent Diagnosis Users commonly encounter four problem categories: remote access failure, start/stop failure, performance anomalies, and disk expansion issues. The diagnosis tool checks system services, disk health, network health, and OS configuration, enabling one‑click health checks.

ECS Automated Repair After diagnosis, automated repair resolves issues within 1–3 minutes, handling system services, network, and disk problems. The process is transparent and compliant, using OOS orchestration and Cloud Assistant commands, with open‑source code, rollback via snapshots, RAM role‑based access control, and audit trails via ActionTrail.

3. AI and Data Behind Unattended Operations

AI + data powers the intelligent diagnosis and automated repair. A data middle‑platform collects, cleans, and analyzes physical, virtual, network, control‑plane, and guest‑OS data, feeding machine‑learning models that generate user profiles, decision trees, prediction, and recommendation models for precise anomaly handling.

The overall ECS self‑service architecture relies on real‑time monitoring (logs, middleware, API, console) and a machine‑learning engine to trigger alerts and drive OOS‑based automated remediation.

Current metrics show over 70% accuracy for real‑time memory anomaly detection and prediction latency under 100 seconds. By integrating expert knowledge, case libraries, and knowledge bases, a robust diagnostic decision tree accelerates issue localization and resolution.

In the past two years, the Elastic Compute team has continuously built an anomaly behavior dataset, planning to evolve it into Alibaba Group’s “ImageNet‑style” dataset for anomaly prediction and open it to the community.