Application‑Based Automated Capacity Management and Utilization Evaluation

Author Introduction Chen Jianming, senior data‑analysis manager at Ctrip Technology Center, focuses on operational data analysis, business order forecasting, and infrastructure capacity management. The content originates from his talk at the GOPS 2016 Global Operations Conference.

1. Introduction The talk addresses automated, application‑oriented capacity management and evaluation, emphasizing that capacity decisions should serve the application’s needs rather than merely keeping servers healthy.

2. Why Resource Utilization Is Low Three main reasons are identified: (1) pursuit of ultra‑fast response times leading to over‑provisioning; (2) developers over‑estimating resource needs as a safety buffer; (3) long procurement cycles causing bulk requests. These practices waste money and energy, with average server utilization reported around 12%.

3. Determining a Safe Utilization Range Based on industry data, utilization below 25% is considered safe, 30%‑40% warrants caution, and exceeding 40% is dangerous and requires immediate scaling. Utilization under 20% is deemed wasteful.

4. Methodology for Improving Utilization The proposed workflow controls traffic distribution by adjusting weights on the front‑end load balancer, directs production traffic to a test server for stress testing, and monitors performance metrics in real time. When a resource reaches a bottleneck, the system records the point and restores normal weights.

5. Data Collection and Analysis After each test, performance data are aggregated to identify the bottleneck and calculate the maximum sustainable request volume (TPS). Multi‑application environments require multivariate regression models to attribute resource impact to each application.

6. Benefits to Developers The approach answers key questions: maximum supported traffic, performance changes after a new release, capacity needed for traffic spikes, and current capacity usage of an application cluster.

7. Dev‑Ops Collaboration With reliable capacity data, ops can trigger automated scaling (VM provisioning and application deployment) within minutes, reducing deployment time from hours to under ten minutes. Cost awareness acts as an implicit control on resource consumption.

8. Prerequisites Effective automation requires comprehensive system‑level and application‑level monitoring to ensure sufficient data for intelligent operations.

Conclusion By integrating automated capacity evaluation, organizations can raise utilization to a reasonable level, guarantee application stability, lower operational costs, and prepare for the upcoming era of intelligent operations.