NetEase Games' AWS Cloud Operations and Practices for Global Game Deployment

Sun Guoliang, who joined NetEase Games in 2013 and initially worked on game operations for titles such as Tiantian and Chu Liuxiang , shares the company's experience of deploying overseas games on AWS, aiming to provide practical guidance for teams new to cloud deployment and to discuss best‑practice optimizations for those already on AWS.

The company began overseas mobile game releases around late 2014, with flagship titles like Onmyoji , Knives Out , and Identity V , and later added newer releases such as LifeAfter and Quantum Agent . Their rapid growth highlighted fundamental differences between domestic and overseas deployments, including the need for public‑cloud infrastructure, virtualized resources, elastic scaling, and global network considerations.

Key challenges identified were performance uncertainty on virtual networks, dynamic elasticity cost management, security (both network and data), and the requirement for global unified servers. To address these, NetEase built a standardized cloud evaluation system that assesses compute, storage, network quality, security, support, and cost factors.

Case studies illustrate concrete issues: a performance discrepancy between Japanese and North American servers of Onmyoji traced to differing clock sources (HPET vs. TSC), and CPU performance loss after Spectre/Meltdown patches affecting the network‑intensive Knives Out service. Solutions involved customizing AWS instance initialization, enabling multi‑queue NICs, activating IRQ/RPS/RFS, and migrating from older instance families to the Nitro‑based C5 series, which offered up to eight network queues and higher utilization.

The evaluation also covered storage (IOPS, throughput, cost‑effectiveness) and network latency between data centers and player regions, using SDK‑embedded probes to measure real‑world latency and packet loss.

Operational practices on AWS rely on core services (EC2, VPC, ELB, S3, CloudFront) with additional security and monitoring tools. The VPC design evolved from simple public/private subnets to a more sophisticated architecture featuring Transit VPCs, VNF gateways/routers, GRE + IPSec tunnels, and later AWS Transit Gateway and cross‑Region peering to reduce latency and cost.

For global service, NetEase routes players to the nearest AWS region (e.g., US West) before forwarding traffic to the primary game servers (e.g., Singapore), leveraging AWS backbone to improve stability. When latency requirements exceed what a single hop can provide, they deploy regional battle servers.

Future plans include exploring new AWS services such as ARM‑based instances, Global Accelerator, VPC TGW, containerization for micro‑service migration, and further cost‑optimization through dynamic scaling and reserved instance strategies.

Overall, the talk emphasizes a continuous loop of requirement‑driven cloud evaluation, solution design, implementation, and iterative improvement, underscoring the importance of practical testing, cost awareness, and ongoing learning in large‑scale game cloud operations.