Professional Cloud ArchitectAnalyze and optimize technical and business processesMedium
A media company is migrating its video transcoding pipeline to Google Cloud. They currently use a queue-based system with on-premises VMs that are often underutilized outside of peak hours, leading to high operational costs. The transcoding jobs are compute-intensive, can be interrupted and restarted, and have flexible deadlines (within 24 hours). They want to optimize costs while ensuring all jobs complete within their deadlines. Which Google Cloud compute option is most cost-effective for this workload?
- AStandard Compute Engine VMs in a Managed Instance Group (MIG).
- BGoogle Kubernetes Engine (GKE) Autopilot with custom node pools.
- CPreemptible VMs (PVMs) with a robust job queue and retry mechanism.
- DCloud Functions for each transcoding task triggered by new video uploads.
Show answer & explanationAnswer & explanation
Correct answer: C. Preemptible VMs (PVMs) with a robust job queue and retry mechanism.
Preemptible VMs are significantly cheaper than standard VMs and are ideal for fault-tolerant, interruptible batch jobs like video transcoding with flexible deadlines. Combining them with a robust job queue and retry mechanism ensures that jobs are eventually completed even if VMs are preempted, providing maximum cost savings.
Why the other options are wrong
- A. Standard Compute Engine VMs are reliable but more expensive than PVMs, leading to higher costs for this interruptible workload.
- B. GKE Autopilot is excellent for containerized workloads and operational simplicity, but PVMs offer more direct cost savings for interruptible compute, and custom node pools don't inherently provide the same cost efficiency as PVMs for this specific use case.
- D. Cloud Functions have execution limits (e.g., memory, time) that might not be suitable for long-running, compute-intensive video transcoding tasks, and their pricing model might not be optimal for sustained, heavy compute.
Preemptible VMs (PVMs)
Compute Engine instances that you can create and run at a much lower price than regular instances, but Google Cloud might terminate (preempt) these instances if it needs the resources elsewhere.
- Up to 80% cheaper than standard VMs.
- Instances last a maximum of 24 hours.
- Ideal for fault-tolerant batch jobs, stateless applications, and dev/test environments.
- Require robust retry logic or checkpointing for reliability.
Memory trick: Cheap, interruptible, retry-able: PVMs for the win!