A Google cloud services guide for the enterprise
A comprehensive collection of articles, videos and more, hand-picked by our editors
Google has rolled out its response to Amazon Web Services' Spot Instance. And while the service can be useful for...
By submitting your personal information, you agree that TechTarget and its partners may contact you regarding relevant content, products and special offers.
a variety of industries, IT pros should be cautious about what they use it for.
Google's Preemptible VMs are now generally available, allowing its customers to use instances at 30% of the standard cost. Those cost savings come with strings attached, however, as customers can be booted at any time and will always be terminated after 24 hours.
The probability that instances will be terminated for a system event is generally low, but varies based on day-to-day conditions and region, Google said. Customers are given 30-seconds notice when instances are terminated and, because of the limitations, should restrict use cases to those that are stateless and don't have strong uptime requirements.
"It's guaranteed at least once every 24 hours, so there are plenty of workloads where this is a great use case and there are also plenty where you may not want to run this," said Jason Stowe, CEO of Cycle Computing LLC, which provides high-performance computing clusters for big data and big compute workloads and has a suite of services available on Google Compute Engine.
Cycle Computing used Preemptible VMs in a partnership with the Broad Institute, a biomedical and genomic research center in Cambridge, Mass., because of the organization's interest in using Google cloud services for scientific workloads, Stowe said.
Jason StoweCEO, Cycle Computing LLC
The Broad Institute needed a significant amount of machine learning to create a map of relations between cellular data sets to assist in its cancer research, and had to set up more than 50,000 core workloads to do so. Preemptible VMs provided cost savings and allowed the Institute to compile and compare its research in a manner that would have taken several months longer using existing internal resources, Stowe said.
Newer applications are better suited for Preemptible VMs, but users have to look at specific applications and how they behave before determining if it's the best fit, he added.
"It's another option for customers and, from a practical standpoint and cost-for-performance [perspective], it enables researchers to get bigger and tackle workloads they might not otherwise be able to," Stowe said.
Preemptible VMs went into beta in May and tens of thousands of jobs have launched since. The service is a good fit for financial services, big data, insurance, manufacturing, life sciences and engineering, Google said.
Cycle Computing also works with Amazon Web Services Spot Instances, and the biggest difference between the two services is the pricing and accessibility models, Stowe said. Spot Instances could run for months without interruption and there are no upper-bounds, unlike Preemptive VMs, which are a finite resource. In addition, the costs for Spot Instances vary, with customers essentially bidding what they're willing to pay, as opposed to the fixed rate for Preemptible VMs, which is 30% of the standard rate per instance type.
But the scalability of both services is comparable, Stowe said.
"Really it's up to the users to benchmark the different types and see where they work best," Stowe said.
There's certainly a niche market for computing power on-demand for low impact workloads, but the users will likely be in high-tech and Web development, said Carl Brooks, analyst with 451 Research LLC, based in New York.
"The folks who use this are definitely not bread-and-butter IT," Brooks said.
Despite Google's warnings about regular interruptions and framing it as disposable resources for customers, it would be surprising if it actually cut people off en mass for its own internal needs, Brooks said.
"I can't really see [Preemptible VMs] being a fundamental part of Google's operational strategy," Brooks said. "It's just a way to recoup losses on infrastructure that would either be taking up space or electricity and not being paid for."
Trevor Jones is a news writer with TechTarget's data center and virtualization media group. Contact him at firstname.lastname@example.org