@PythonHub Burla uses Ray-style distributed compute primitives. 1000-node scaling needs efficient task serialization, low-latency scheduler. Reduced network overhead by 50% vs naive RPC. Running on Kubernetes or bare metal for pod orchestration?