GAMEPERF05-BP01 Benchmark your game performance across multiple compute types

For game server workloads, there is no singular approach to identifying the optimal compute solution for hosting your game server. A common strategy for benchmarking game servers is to start with compute-optimized EC2 'c' instances, because this instance family provides high performance for workloads that are computationally intensive. Alternatively, if your game requires a significant amount of memory to implement specific features, the memory-optimized instances may be most suitable.

Level of risk exposed if this best practice is not established: High

Implementation guidance

If your workload utilizes significant network resources, consider implementing instances that are network-optimized which is typically indicated using an 'n' in the instance name, avoid burstable instance types 't' as after credits are exhausted performance will decrease. Games are sensitive to latency and dropped packets, so it is recommended to use EC2 enhanced networking to help improve the network performance of your game servers. Enhanced networking uses single root I/O virtualization (SR-IOV) to provide high-performance networking capabilities on supported instance types . SR-IOV is a method of device virtualization that provides higher I/O performance and lower CPU utilization when compared to traditional virtualized network interfaces. Enhanced networking provides higher bandwidth, higher packet per second (PPS) performance, and consistently lower inter-instance latencies. Enhanced networking with Elastic Network Adapter is available for most recent EC2 instance types and is important to regularly update to benefit from performance enhancements from newer instances and improvements to the AWS Nitro hypervisor.

If your game performs similarly across multiple EC2 instance types, then you should consider using multiple instance types to host your game servers. Monitor performance over time and perform further optimization after you have hosted enough production game sessions to be able to identify performance trends. Remember that your compute requirements may change as you add new features into your game that require different allocation of resources. You can configure EC2 Auto Scaling groups to use multiple instance types, or you can use separate Auto Scaling groups to host game server instances that run separate instance types which may make it simpler to manage correlation and aggregation of metrics. 

Evaluate how your game performs on different types of processors such as Intel-based instances, AMD-based instances, and ARM-based Graviton instances. Unreal Engine 5.1.1 or newer can compile game servers for Graviton and can improve price performance for your game. Perform sweep and saturation testing at various sizes within each family to determine the sweet spot where utilization and performance are consistent.

You should also benchmark how your game performance is impacted when it is hosted using containers and Lambda functions. For use cases where long-lived game server processes are not required, such as asynchronous games and for game backend services, consider using a serverless architecture with Lambda which can simplify management and operations for game operations teams, as well as allow you to more quickly deploy your game globally to many AWS Regions. For serverless best practices, refer to the Serverless Applications Lens - Well-Architected Framework.

Implementation steps

For more information, see Choose the right compute strategy for your global game servers.