Back in May, Google announced that it would be offering its Compute Engine VMs in a new flavor. Called preemptible VMs (presumably as opposed to contemptible VMs which would be far less desirable), these compute instances are priced at rock bottom — at a 70% discount to regular instances.
There is, however, a catch. Preemptible VMs can be shut down at any time and are limited to a 24-hour runtime. To add to the uncertainty, users get only 30 seconds' notice of a VM shutdown. Ouch.
Of course there is a method to the madness and certain workloads lend themselves to this level of transience in terms of computing. A situation where a huge compute job needs to be done, but isn't time critical, and has an operational management layer sitting above it, for example. Another mooted suggestion is to use these VMs to enhance performance while running the main part of an application on conventional VMs to maintain uptime. Load or security testing look like another useful use case for these most temporal of instances.
Preemptible VMs obviously compete with spot instances from Amazon Web Services (AWS), but, as is usually the case, the two companies differ slightly in the way they approach these cut-price instances. With AWS, users name their own price for an instance — the higher price they bid, the less likely it is that their instance will be interrupted.
Google markets these VMs as a new class of computing. The jury is out on whether that is the case. It is true that there are differences between AWS's and Google's approached but whether that is enough to justify calling it a new class of cloud is another thing entirely.
Along with the general availability announcement, Google is also announcing that Cycle Computing's full suite of products will "soon" work on Google Compute Engine. Cycle Computing specializes in managing and running massive workloads for science and research-based customers. Hence its ability to use preemptible VMs is useful.
Cycle Computing recently ran a 50,000-plus core workload for the Broad Institute using its CycleCloud with Google preemptible VMs. The results were impressive; using the tools, the institute performed the equivalent of three decades of cancer research computations in an afternoon.
“These types of analyses provide the clues that can lead to breakthroughs in disease research, such as cancer, and this kind of cloud-based infrastructure helps us remove some of the local computing barriers that can stand in the way,” said Chris Dwan, Acting Director of Information Technology at the Broad Institute. “Flexible processing power allows us to think on a much larger scale.”
According to Cycle Computing, customers already using CycleCloud will find that Google Compute Engine Preemptible VMs are an easy choice for reducing costs for financial, big data, insurance, manufacturing, life sciences and engineering workloads, regardless of the number of cores needed. “Preemptible VMs are a great new offering for big compute, analytics, and batch workloads, and we’re happy to support them as an option for our customers. The Broad Institute’s story showcases a production workload running in 6 hours instead of 6 weeks, thanks to preemptible VMs”, said Jason Stowe, CEO of Cycle Computing.
Preemptible VMs are generally available from Google today.