Is there a spot instance-like mechanism with google cloud? I regularly request high memory spot instances (r3.8xlarge) on AWS for 1-2 days processes (genomic analysis). With spot-pricing it can be pretty cheap.
A very clever way to run genomic analysis on GCP is via WDL ( https://software.broadinstitute.org/wdl/ ) workflows executed on the Cromwell ( https://github.com/broadinstitute/cromwell ) service. Using WDL you define tasks, and compose those into workflows, then Cromwell executes those tasks on GCP using the pipelines api ( https://cloud.google.com/genomics/reference/rest/v1alpha2/pi... ). So you don't need to set up any instances at all, each stage of your pipeline, a task, is executed in its own docker instance, with files from google cloud storage being materialized on that instance for you. When that task is done, outputs are sent to cloud storage buckets, and everything is torn down.
Picard can pull reads directly out of the google genomics API, and more of the Broad tools are adopting the GA4GH api, so this means an increasing amount of core tools anyone doing genomics will want to use, will work directly with google genomics.
Preemptible Instances only can be used for a max of 24 hours, so would not fit the above usage pattern...though I don't know if AWS spot instances are really meant to be used for that long either. Of course both allow you to use the regular non-preemptible on-demand pricing.
