Request to add larger ensemble to CESM testing

[raeder]

I ran into a memory problem using CESM1_5_beta03, and a compset which includes CISM1(5 km resolution), which is single threaded.  There's a single-threaded version of CISM2 available too, which would have the same problem.  I set up a job with 15 instances, and the only available(?)layout of the CISM instances put them all on 1 node.  This overwhelmed the memory.  I had to change to a 20 km resolution in order for this job to work.  This will probably be fixed in the parallel version of CISM2, but if that won't be available for a whileit would be helpful to have some other fix.  I see variables like PSTRID_GLC in the CESM1_3 namelistfor env_mach_pes.xml, but it can/could only be set to 1.  If it could be set larger, then the CISM1 instancescould be spread out over more nodes.Related to this, it would be helpful if the standard CESM testing included a short, multi-instance test with at least15 instances, so that problems like this could be discovered early.Thanks for considering,Kevin Raeder

 

[l_vankampenhout@uu_nl]

Kevin,CISM2 supports the 'Glissade' dynamical core, which is parallel in all the available approximations (including SIA). Please refer to the CISM documentation at http://oceans11.lanl.gov/cism/documentation.htmlpage 102 and further, e.g. options dycore, which_ho_approx    Best, Leo

 

[raeder]

Leo,thanks for the comment and link!   That is definitely the preferred solution, but in the short term may requirebuilding and using new compsets.  It seems like the CESM developers intend to implement the ability tospread instances out over multiple nodes, and I'm hoping that this will prompt them to do that.