Contact us

Join the community

Scan with WeChat
Join the official community group

Try Guance

Start online with usage-based pricing and a true cloud service.

Get started

Choose a Guance edition

Code repositories

Heroes entertain each other Guance

In high-concurrency scenarios, gaming performance is improved by more than 3 times

Improved application performance
Code tuning
Contact us

Customer background

Hero Entertainment is a global leader in interactive entertainment, dedicated to providing high-quality mobile gaming experiences to users worldwide. Many of its games are popular among the public, and while their popularity continues to rise, they also face the following challenges:

For some potentially hit games, there are cases where a large number of login requests suddenly appear. To better address this scenario, customers are trying to deploy services containerized and leverage HPA (Horizontal Pod Autoscaling) capabilities to improve system capacity through rapid elastic automatic scaling during sudden surge in requests.

The Kubernetes cluster on the public cloud is used, aiming to enable business-related services to instantly reach 5W RPS capacity through the automatic elastic scaling of container services. Stress testing showed that with 6 pods, the RPS peak load was around 2W. Increasing to 12 Pods (same resource configuration), RPS reached a peak load of about 2.4W, and continued scaling performance barely improved.

Solutions

Deploy DataKit in a container service environment to access link and log data.

During stress testing, by Guance the APM application performance monitoring service overview page, you can quickly locate services and related resources with suspected performance bottlenecks. Based on Redis analysis on the cloud, requests for Redis usage have basically reached the peak capacity supported by configuration specifications.

Enable profile collection in the Guance and perform performance analysis in conjunction with the application link, focusing mainly on the "Lock Wait Time" and "Socket I/O Read Time" metrics, thereby locating the code to frequently call Redis through different implementation methods to complete related operations. Feedback is provided to R&D for code optimization.

Customer outcomes

After the code iteration went live in the new version, in the latest round of stress testing, without changing the original resource allocation, the peak RPS reached around 80,000, exceeding the client's original expectations.

More cases