What is the input/output bandwidth limits for ECS Fargate tasks?

The network bandwidth limit is something which is not published as we cannot be deterministic about it. In general, more vCPU should get you greater network bandwidth

The network bandwidth limits for Amazon ECS Fargate tasks depend on the number of vCPUs allocated to the task. This information is provided in the AWS documentation on "Performance of Amazon EBS volumes for Fargate on-demand tasks".

According to that document, the baseline bandwidth limits for Fargate tasks are:

• 2 vCPUs: 360 Mbps
• 4 vCPUs: 1,150 Mbps
• 8 vCPUs: 2,300 Mbps
• 16 vCPUs: 4,500 Mbps

The excerpt from the documentation states:

"If you request more than 1 vCPU for your Fargate task, the following baseline performance limits apply to an Amazon EBS volume attached to the task...

CPU units requested (in vCPUs) | Baseline bandwidth (in Mbps) 2 | 360
4 | 1,150 8 | 2,300 16 | 4,500"

So the network bandwidth available to your Fargate task scales linearly with the number of vCPUs you allocate, up to a maximum of 4,500 Mbps for 16 vCPU tasks. These are the baseline bandwidth limits you should plan your workloads around, though you may see higher bandwidth temporarily. Sources [1] [Performance of Amazon EBS volumes for Fargate on-demand tasks - Amazon Elastic Container Service] (https://docs.aws.amazon.com/AmazonECS/latest/developerguide/ebs-fargate-performance-limits.html)

[2] [Managing your Amazon ECS and AWS Fargate service quotas in the AWS Management Console - Amazon Elastic Container Service] (https://docs.aws.amazon.com/AmazonECS/latest/developerguide/service-quotas-manage.html)

[3] [Viewing recommendations for Amazon ECS services on Fargate - AWS Compute Optimizer] (https://docs.aws.amazon.com/compute-optimizer/latest/ug/view-ecs-recommendations.html)

Per the article you listed above, you can also do some benchmark testing yourself with an ec2 instance and iperf to determine the upper limit of what you can expect. You can tune your iperf command to simulate the type of traffic you would expect to throw at your ECS Fargate application to see how the infrastructure will handle it and scale accordingly. In this case, you would do that by adding additional vCPU.