You manage a GKE cluster that runs a mix of critical web services and fault-tolerant batch processing jobs. The CFO has asked you to drastically reduce the compute costs of the cluster. How should you optimize the cluster architecture?

GKE allows you to create multiple node pools. By scheduling the fault-tolerant batch jobs onto a node pool backed by Spot VMs (formerly Preemptible VMs), you save massive amounts of money. Keeping the critical web services on standard VMs ensures reliability. Node Auto-Provisioning further optimizes costs by creating nodes that perfectly fit the resource requests of your pods.