BISMG:Monsoon

Back of envelope calculation suggested that glimmer-cism might want to use around 1GB of RAM, possibly comparable to HadCM3. DAGW and RMG established that using 'allocatable' we can allocate at least 8GB on either monsoon or BC2 head nodes. We anticipate this might be higher on the Monsoon computational nodes (perhaps up to 64GB), but the same on the BC2 computational nodes. In either case, it looks like we have sufficient RAM available.

Actions/active discussions


 * Currently only compiler available is xl (an IBM compiler). DAGW and RMG currently discussing this.


 * RG to try to install glimmer-cism-lanl on Monsoon using xl.

Notes on this so far:

Am trying to install on the fast disk on the hpc machine, ibm00: /projects/icesheet/ruglad/glimmer-cism-lanl

A newer automake is required by the gcl bootstrap script. I've hacked this for the time being - will see what happens. The only available netcdf lib doesn't appear to be compatible with the fortran compiler...


 * Currently only UM version 4.7 is installed on Monsoon. We use UM 4.5 for the ice sheet coupling, it is not clear how much work porting the coupling would be.  It is possible that 4.5 will be installed by Lois Steenman Clark.  Alternatively we will need to either install 4.5 ourselves or modify our modset.

- We do know that both UMv4.5 and glimmer-cism compile with ifort. However, the XL compiler is likely to give us faster executables.

- We don't know whether we need to use UMv4.5 on Monsoon yet.


 * RMG and DAGW need to find out how to submit parallel job other than via UMUI (might be on Twiki, will look when back up).


 * DAGW and RMG to investigate port forwarding as a way to allow easier Monsoon connectivity


 * We can't use svn, as no ssh out. DAGW can write an rsync script so that we can keep our checkouts on e.g. dartagnan in sync with those on monsoon, if needed.


 * The module command doesn't seem to work in a bash shell on Monsoon, RMG to inquire.