Performance regression after 20250909 x86

[clemstoquart]

Hello,

We have a CPU and I/O intensive Node.js workload which shows a 30% throughput regression with all AMIs starting with this release. CPU consumption is lower and throughput is also way lower.

I'm having a hard time finding why because differences between 20250909 and 20250916 seem minimal.

Do you have any idea about a change that might cause such a big difference between these two AMIs ?

[harishxr]

HI @clemstoquart ,

Assuming you are talking about the AL2023 ECS Optimized AMD64 variant of the AMI, the base image that we build upon has changed from al2023-ami-minimal-2023.8.20250818.0-kernel-6.1-x86_64 to al2023-ami-minimal-2023.8.20250908.0-kernel-6.1-x86_64

These are the changes that have gone into the Base AMI with that particular release https://docs.aws.amazon.com/linux/al2023/release-notes/relnotes-2023.8.20250908.html#amis-2023.8.20250908.Minimal-Kernel-6-1-AMI

Can you give us some more information on your workload? Are you running your workload as an ECS task? If yes, could you give us some more information about your workload, your task definition, and also agent logs using https://github.com/aws/amazon-ecs-ami?tab=readme-ov-file#ecs-logs-collector.

Please send this information over to ecs-agent-external@amazon.com

[clemstoquart]

Hello,

Yes, our workload is running as an ECS task. It's an async workload that reads and publishes messages from SQS, decrypts data using KMS and them performs a lot of HTTP calls to external services outside of our VPC.

Thanks for the ecs-logs-collector, I'm going to have a look at it.