Dataproc is pretty significant compared with other managed Hadoop/Spark services. Here's how: - You get a cluster in under 90 seconds and with per-minute billi

Dataproc is pretty significant compared with other managed Hadoop/Spark services. Here's how: - You get a cluster in under 90 seconds and with per-minute billing. - This lets you start with a job, then start a Dataproc cluster as part of job execution, then save output and discard cluster - You aren't throttled by slow startup times, and you are not worried about rounding up your workloads to nearest hour Plus Preemptible VMs support at 70% off, Custom Machines types, GCS as storage layer, very very fast networking, and interoperability with other GCP services. (disclosure: work on GCP but not Dataproc)