Open delagoya opened 5 years ago
I agree on using MiB. Besides the reason you state, we recently encountered some issues in Cromwell as users were making hyperoptimized WDLs and running afoul of things as they didn't understand the difference (WDL supports both units). What we wound up doing was always treating MB as MiB (and so forth up the unit chain) as for those types of users that's usually what they meant anyways, and MiB > MB so it was the safer fallback
AWS Batch, Azure Kubernetes, and GCP all seem to use MiB for memory definitions. They also use
integer
instead ofnumber
Propose we move to this model for defining Task memory requirements over a
double
GB.