On-demand high performance compute (HPC) can help to solve problems ranging from medical imaging, genomics, visualization, and machine learning. By leveraging the “spare” cloud capacity of major cloud vendors, costs and time required to solve difficult problems can be significantly reduced.
Google announced this week that preemptible virtual machines are now generally available: https://cloud.google.com/compute/docs/instances/preemptible. They first offered the product in beta this spring.
Amazon Web Services (AWS) has offered Spot Instances since 2009 and recently launched “Spot Fleet” which launches a target number of spot instances and places them in the lowest priced availability zone.
Spot instances and preemptible machines may not always be available for consumption depending on demand. Often may be more available at night or weekends for example.
Because Google and Amazon may terminate these machines at any time, they are not typically suitable for workloads on them that are not fault-tolerant.
For the kind of distributed compute and batch workloads that can failover when one machine goes down, though, preemptible and spot-instance machines can offer significant cost savings.
Google Compute Engine Preemptible Virtual Machines, are the same as regular instances except for a few key differences – they are low cost and they may be shut (preempted) down at any time.
– fixed pricing
– 30 second notice before termination
– 24 hour runtime limit
– pricing is fixed at approximately 1/2 of standard VM https://cloud.google.com/compute/#pricing
Groups like CycleComputing can help users leverage and orchestrate the power of large numbers of preemptible instances.
Amazon Web Services Spot Instances enable you to bid on unused EC2 instances, which can lower your Amazon EC2 costs significantly. The hourly price for a Spot instance (of each instance type in each Availability Zone) is set by Amazon EC2, and fluctuates depending on the supply of and demand for Spot instances.
– variable bid/pricing
– operate in public or virtual private cloud (VPC)
– spot market pricing by geography http://docs.aws.amazon.com/AWSEC2/latest/UserGuide/using-spot-instances-history.html
Amazon acquired startup ClusterK in May 2015 to provide ready access to spot instances and advanced HPC grid scheduling.
The Bottom Line
- For many workloads, customers should be optimizing their costs to leverage spot/preemptible instances.
- Cost savings using these interruptable instances can be significant.
- Fortune Article – Cheap Spot Instances
- Cycle Computing and the Broad leverage Preemptible Instances for cancer – article
