Packing More RL Workloads: Firecracker Density Tuning on EC2 Metal

Running reinforcement learning at scale means spinning up thousands of isolated sandboxes in parallel—and Firecracker microVMs on Amazon EC2 metal let you cram way more of them per host, cutting your cost per training step. This post walks through tuning microVM density and introduces labsweep, a tool for plotting your own latency-versus-density tradeoffs.