Azure: Devito-MPI strong scaling POC

[ggorman]

Initial objective is to rerun same model we were running on cx2.

• Set up blob storage for input files rather than setting up a parallel FS.
• Processor type is not critical initially.
• Required: Infiniband interconnect.
• Required: Optimized MPI distribution.

[georgebisbas]

Status until now:

• We have already executed a few benchmark trial runs on Azure.
• Machines used are H series, can go up to 12 vms, just submitetd request to 10x our resources.
• Infiniband is supposed to be provided. Should I test it somehow?

[ggorman]

Note:
#8 creates a recipe is to run an MPI benchmark across N MPI processes on X architecture.
devitocodes/devito#1089 takes the recipe from #8 executes it for different values of N (and X when we get that far).