Algorithm adjustments, improved efficiency, and usage pattern analysis has enabled Google to slash the price of its Preemptible VMs by up to 33 percent, the company announced Tuesday. Preemptible VM instances can now be used for a penny an hour, or 80 percent less than the equivalent non-preemptible VM on Google Cloud Platform.
Preemptible VMs are basically Google Compute Engine VMs that the company needs to have in reserve for spikes in demand, but will rent out in blocks up to 24 hours, with the condition that they can be preempted if a spike occurs.
The AWS equivalent is EC2 Spot Instances, which it auctions off based on capacity. AWS acquired infrastructure optimizer ClusterK in 2015 to help run Spot Instances efficiently.
While not very helpful for most cloud workloads, utilizing excess VMs to run workloads without major time constraints could significantly reduce computing cost.
“Our customers are using Preemptible VMs to analyze data, render movies, process satellite imagery,analyze genomic data, transcode media and complete a variety of business and engineering tasks, using thousands of Preemptible VM cores in a single job,” Google Product Manager Michael Basilyan said in a blog post. “We believe that the price reduction for Preemptible VMs will unlock even more computing opportunities and enable you to tackle interesting science and business problems.”
Google pre-empts more recently launched instances first, to cut down on lost work, and VMs pre-empted in the first 10 minutes are not billed. Preemptible VMs can be launched in several different ways, which are outlined in the blog post, along with tips to get the most out of the service.
Unlike many previous cloud price cuts, the niche market and technical controls necessary make it unlikely this change will be copied by other public cloud providers, with the possible exception that Azure may begin offering a discounted way to purchase non-guaranteed VMs in the future.