The smaller and cheaper EC2 instances are listed as having network performance using the phrase "up to". For the longest thing, I always thought this meant that performance would match that number but might possibly drop temporarily.
After doing some research, I came across this Twitter thread by @dvassallo. Apparently, my misunderstanding is much more common than I thought. There appears to actually be a credit system of sorts and the listed "up to" value is actually best case scenario. Once best case usage is sustained for too long, your instance will have its network performance limited.
In the same Twitter thread, this blog post is referenced which agrees with Daniel's research.
I've had situations come up at work where network performance would suddenly get worse and it was never clear why that happened until now.
Both listed resources have excel sheets with some numbers, but I wanted to throw together a website that made this information even more accessible and indexable. So that's what I'm doing! The name of the website: ec2throughput.info is taking direct inspiration from ec2instances.info.
I'm using a combination of Pulumi's infrastructure as code solution (all under the ./operations) combined with GitHub Actions to apply the Pulumi code. I run iperf3 on two machines that are placed in the same availability zone in AWS. These tests run for an hour and we store the information in S3.
The website at https://ec2throughput.info is hosted on Vercel. Whenever a new iperf3 test is complete, the build process is kicked off where the website bundle will download the latest data we have for each instance type and will convert that into a static, client side only website.