You run nightly batch jobs that take ~2 hours and want to minimize costs for these noninteractive jobs. Which compute resource choice should you use?

Preemptible VM instances are ideal for fault-tolerant, non-interactive batch jobs because they offer significant cost savings (up to 80% compared to standard instances). While they can be terminated by Compute Engine if resources are needed elsewhere, the 2-hour duration of your batch jobs makes them a good fit, as you can design your application to handle preemption or restart. Google Kubernetes Engine (GKE), whether with a single-node small cluster or a three-node cluster of micro instances, introduces orchestration overhead and continuous resource consumption that might be overkill and more expensive for simple, nightly batch jobs compared to preemptible VMs. GKE is generally better suited for long-running, containerized applications requiring high availability and scalability. Compute Engine instance types that support micro bursting are not a standard offering and don't directly address cost minimization for batch jobs in the way preemptible VMs do.