Google Preemptible VMs

Saving 80% on Google Cloud Compute Costs

Google offers compute capacity at a lower cost than on-demand instances, with preemptible VMs.  Preemptible VMs are excess compute capacity that can help companies optimize their cloud spend and reduce costs by 80%.

What’s the difference between preemptible VMs and on-demand?

The instances you purchase from GCP are the same, regardless of whether they are bought as preemptible, on-demand or as a long-term commitment. Instances of all types, sizes and from all regions are available across the compute pricing models. Nor do users have to bid on preemptible VMs, as Google has set a fixed price for them just like standard instances. Pricing details for all can be found in Google Cloud documentation.

What makes it possible for GCP to offer preemptible VMs at such a low cost—up to 80% lower than on-demand—is that they can be terminated with little warning when GCP needs to reallocate the compute elsewhere. Because of this, these short-lived, preemptible VMs have no availability guarantees. Other things to note about preemptible instances:

  • After running for 24 hours, a preemptible VM is always terminated, unless the instance has been reset (e.g. stopped and started)
  • When Google Cloud needs to reclaim capacity, termination can occur with only a 30 second warning
  • As preemptible VMs are excess capacity, they may not always be available
  • New instances are preempted over older instances

How do preemptible VMs work?

Because of the risk of termination, Google recommends that users run fault-tolerant workloads and batch jobs on preemptible VMs. But with no SLA , you can achieve better availability with preemptible VMs when you diversify instance types to minimize the risk of losing a bunch of VMs at once.

For workloads that require any availability, consistency or data integrity, managing these workloads on preemptible VMs requires extra configuration and management. In general, running simulated maintenance events to force instance to preempt is a good way to test and understand how your applications and workloads will handle preemptible VMs. Learn more about running test maintenance events on your instances.

Customers can also leverage both on-demand and preemptible VMs in their long-running clusters, using a combination of different kinds of compute to achieve maximum efficiency and optimization. Spot by NetApp offers solutions for legacy and containerized applications running on GCP that will enable you to confidently leverage preemptible VMs with high availability.

To make Google Cloud Preemptible VMs a real part of your cloud cost optimization strategy, learn more about Spot by NetApps solutions.