Run EMS without MPITCH?

[mgd08]

Hello,

I am a graduate student at Florida State University, and I'm using the universities "high performance computing" or HPC to run WRFEMS/UEMS. Their HPC already has SLURM installed as the workload manager. This has been causing some issues since MPICH is installed in EMS. I am currently trying to write code that pastes the "hostname1:np1:IP1,hostname2:np2:IP2" that SLURM decides to use as the value for REAL_NODECPUS in the run_ncpus.conf file. However, I was wondering if there is possibly a better way to go about this whole process by some how turnining off MPICH, since I was able to run WRF ARW on our HPC with no issues?

If you are wondering why I am not just using WRF ARW its because the research I am doing is for a region that uses EMS and more importantly uses SPoRTs data (which I don't know how to incorporate into ARW)

Also currently I have 5. HYDRA_IFACE in the run_ncpus.conf file configured as "HYDRA_IFACE = eth0, eth1, eth2, eth3, lo", as it said to run "ifconfig" command and to take note of the names on the left hand side. Since those four different names popped up I figured I should include them all but don't know if this is allowed.

[Antonix]

Hi!
there are two problems to solve
1) slurm is very difficult to use with EMS, I tried it for one month, but he failed. If you can let me know how you did!

2) if you work on a HPC cluster, are probably in use the infiniband network. MPICH2 compiled into UEMS package does not support infiniband network. therefore, does not exploit all of the available bandwidth in infiniband (I think 40 gb/s) but you stop to 1 gb/s.

I suggest you to compile native WRF, I can help them without any problem.

Obviously if you can find solutions to this problem please let us know!