Simulating a large fleet of server hosts for testing purposes

I want to simulate a large fleet of hosts (around 10,000 hosts) behind a load balancer handling requests. The simulated server is be doing significantly lower amount of work than the actual server. Basically, I want to test how a single component in the service request path would behave in a very large fleet. What are some of the good ways to setup this simulation test bed? I am trying to optimize the costs and HW used to build this test bed.

The simulation might run for about 1hr.

Some Ideas I had so far are:

1. Spinning up a fraction of actual host count and running multiple threads each thread representing a single server.
2. Following same approach as #1 but using containers.
3. AWS ECS Fargate to spin up tasks where each task is again simulates multiple servers.

I guess all three options follow the same trend of using threads to simulate servers.