One of the many reasons I have been so quiet of late is that I have been bevering away on a new ENES1 infrastructure strategy for climate modelling. We now have a draft out for public feedback, so please do comment!
There are seven headline recommendations:
- The HPC community must continue to provide both CPU and GPU machines, a lot of climate codes will not be sensibly deployed on GPU machine in the near future (even if they can be made to run on them).
- There is a need for a more operational aspect to some aspects of climate science.
- The community should continue to invest in managing and sustaining shared infrastructure.
- Model development takes a long time and is resource intensive. Modellers will have to pay attention to the choices between Performance, Portability, and (scientific) Productivity; in the new world, we can only have two!
- Large expensive modelling projects need to be treated like satellite missions, well publicised and documented.
- The community should continue to invest in the necessary underpinning diagnostic tools and libraries.
- Storage and data systems need to support a variety of use-cases.2
However, there are 37 numbered detailed recommendations, seven of which target the HPC community - the other thirty target the climate modelling community. There ought to be no surprises in this document, but nonetheless, community feedback would be valuable, especially if there are signficant things missing, or some things need more emphasis.
There is an underlying strategic science case which I am preparing for publication at the moment, you’ll hear more about that soon. You can see some hints of that in a couple of my recent talks (e.g. balajifest and my NCAS seminar).