AWS offers EC2 instances in different sizes defined by instance types. To decide which instance type to use, consider factors like memory needed by OS and app, benefit from multiple CPU cores, and expected network throughput. Information about memory and CPU capacity of instance types can be found at Amazon EC2 Instance Types. For example, an m4.large instance has 2 vCPUs and 8 GiB memory, but AWS only promises "Moderate" network performance without specific GBit/s details. The author has been dissatisfied with the lack of actual network performance info and thus measured EC2 instance networking performance in the real world over the past few weeks.

Summary of benchmarking infrastructure: benchmarking tool is iperf3, instance type of iperf3 server is m4.16xlarge, regions are eu-west-1 and us-west-2, and period is every 3 hours from November 22nd, 2017 to November 28th, 2017. Expect a detailed blog post later.

For the m3 instance family, benchmark results are very consistent with minimal variance. Network throughput is expected to be from 0.3 Gbps to 1.0 Gbps.

For the m4 instance family, variance is higher between measurements of the same instance type. AWS details network throughput for m4.10xlarge (10 Gbps) and m4.16xlarge (25 Gbps).

In conclusion, when choosing an instance type, network performance should be a main criterion, especially for network-intensive workloads. As AWS provides vague network performance info, the benchmark helps avoid bottlenecks caused by instance networking capacity. Networking benchmarks of other instance families will follow.