The statistical computing environment meets the cluster- Matt Bonyak | Domino

Pharma R&D has HPC too, under other names. Trial simulations and sensitivity analyses with thousands of replicates run on a Slurm cluster next door to a GxP statistical computing environment, with scp as the integration between them. When a regulator asks which replicates produced the number in the submission, someone goes looking.

This session shows an R simulation study parallelized with future and mirai, fanned out to Slurm workers from the same place the statistician writes the analysis. The code is the same as what ran on a laptop; only the backend changes. Every replicate is tracked as a job, the results land in versioned shared storage with a NetApp snapshot taken at the point of analysis, and the software environment and code revision are captured with them. Compute spins up for the study and releases when it ends. When the regulator asks which replicates produced the number, you look it up.