#managedinstances — Public Fediverse posts

Last year, I co-presented a "Serverless vs Kubernetes" talk with Davide de Paolis. I took the Kubernetes side, and I got lucky: if AWS had already announced managed instances, the serverless side would have had a much stronger hand.

At that time, we ended in a tie: both serve different needs and different scenarios, with no clear winner.

For years, every conference talk about compute ended the same way: "it depends on your workload." Lambda was brilliant but expensive at scale, Fargate was simple, but you couldn't pick your instance type, while EC2 gave you control, but your ops team aged 10 years maintaining it.

Then Lambda Managed Instances, ECS Managed Instances, and EKS Auto Mode all dropped within months. The idea: EC2 economics with serverless operations. AWS handles provisioning, patching, and scaling. You choose the instance type, apply Savings Plans, and stop babysitting AMIs.

We ran the numbers on a 50M requests/month workload. Standard Lambda: $343/month. ECS Managed Instances with a 1-year Savings Plan: $117. Same workload, same reliability, 66% cheaper.
There's a catch, of course. Lambda Managed Instances scale on CPU utilization, not invocation rate. If traffic doubles in 5 minutes, expect throttling. And ECS Managed Instances pack multiple tasks per instance, so if you need Fargate-level isolation, you'll want single-task mode.

You can find an in-depth comparison with cost breakdowns for steady and bursty traffic, plus a decision matrix for when to use what here: https://blog.besharp.it/when-serverless-runs-on-servers-new-options-for-lambda-and-fargate-with-managed-instances/

#AWS #Lambda #Fargate #ECS #EKS #Serverless #CloudArchitecture #CostOptimization #ManagedInstances