Mastering FinOps in Cloud‑Native Container Environments: Real‑World Practices and Cost‑Saving Strategies

Case Overview

The featured enterprise is a top Chinese quantitative investment company that relies heavily on AI, machine learning, and big‑data workloads for trading decisions. It runs flexible, high‑performance compute jobs in a shared Kubernetes cluster using Alibaba Cloud Container Service (ACK).

Key Challenges

Planning difficulty : Accurately sizing resources for on‑demand tasks and test environments is hard; over‑provisioning wastes resources while under‑provisioning harms stability.

Cost allocation difficulty : Multiple applications share a single K8s cluster, making it hard to map billing to specific apps, teams, or individuals across time and space dimensions.

Management difficulty : Reducing idle resources without compromising service reliability requires coordinated actions across teams.

Optimization difficulty : Adjusting requests, limits, and HPA policies must be done carefully to avoid performance regressions.

FinOps Process Adopted

The organization followed a three‑stage FinOps workflow: cost insight, cost optimization, and cost control.

Cost Insight

ACK FinOps suite delivers granular billing data, including daily per‑resource costs, per‑pod usage, and resource‑level watermarks. By labeling namespaces and workloads, the team built multiple dashboards that map costs to clusters, nodes, applications, and even individual users.

Cost Optimization

Features such as resource profiling, intelligent request/limit recommendations, Koordinator‑based mixed‑tenant optimization, and automated elastic strategies (CronHPA, AHPA) were used. The suite also provides waste inspection to surface deterministic idle resources.

Cost Control

Weekly cost‑report dashboards are automatically emailed to responsible teams. Budget alerts, quota enforcement, and predictive cost trends help enforce spending limits.

Implementation Details

Responsibility was split among three groups: the Infra team (provides infrastructure and drives cost governance), the Business Platform team (sets budgets and quotas), and the Application team (optimizes workloads based on cost insights).

Resource planning leverages ACK AHPA for dynamic scaling, quota settings based on historical usage, and label‑driven cost attribution to departments and individuals.

Waste Identification

Deterministic waste : Unused SLB, EIP, idle ECS instances, and empty nodes are detected and reclaimed.

Application‑level waste : Long‑tail small apps consume disproportionate resources; profiling data maps this waste back to specific pods and owners.

Quota & Request/Limit Configuration

Experienced K8s users set appropriate request and limit values. ACK’s profiling engine suggests optimal values based on multi‑dimensional usage statistics, half‑life weighting, and container runtime health signals.

Elastic Strategies

Standard HPA handles bursty workloads, VPA adjusts node‑level resources, CronHPA schedules periodic scaling, Keda reacts to event‑driven metrics, and Knative supports serverless scenarios. The new AHPA adds predictive pre‑warming, automatic threshold tuning, and downgrade protection to avoid both over‑ and under‑scaling.

Results

After more than a year of continuous FinOps practice, the company saved roughly 25% of its IT spend (over 100,000 CNY per month). Cluster utilization rose from about 20% to 50%, and resource turnover efficiency improved by 20%.

The team hopes this sharing helps other cloud‑native customers build their own FinOps frameworks.