Difference between revisions of "ATMOS:WRF"
Jump to navigation
Jump to search
en>Peterj |
m (1 revision imported) |
(No difference)
|
Revision as of 17:10, 21 August 2020
Set up WRF on rocks.unbc.ca
After installing WRF (EMS), it is necessary to setup the user environment so that the cluster can work:
First, make sure that passwordless ssh works between the head node and all of the other nodes of the cluster (excerpted from here)
- Generate a public/private RSA key pair on a host of your choice, whose home directory is shared with the remote hosts that you want to run jobs on, and enter a blank passphrase when prompted:
cd ~/.ssh ssh-keygen -t rsa
- Add your public key to the list of keys allowed to log in to your account:
cat id_rsa.pub >> authorized_keys cat id_rsa.pub >> authorized_keys2
- Edit the .ssh/config file so that it includes the following:
StrictHostKeyChecking no Host compute-0-0-ib HostName 172.16.0.2 Host compute-0-1-ib HostName 172.16.0.3 Host compute-0-2-ib HostName 172.16.0.4 Host compute-0-3-ib HostName 172.16.0.5 Host compute-0-4-ib HostName 172.16.0.6 Host compute-0-5-ib HostName 172.16.0.7 Host compute-0-6-ib HostName 172.16.0.8 Host compute-0-7-ib HostName 172.16.0.9 Host compute-0-8-ib HostName 172.16.0.10 Host compute-0-9-ib HostName 172.16.0.11 Host rocks-ib HostName 172.16.0.12
Next, test that WRF recognizes the network:
mpicheck rocks 172.16.0.2 172.16.0.3 172.16.0.4 172.16.0.5 172.16.0.6 172.16.0.7 172.16.0.8 172.16.0.9 172.16.0.10 172.16.0.11
Next, edit the file run_ncpus.conf so that the network is defined, note that we are using 1 less physical core than exists on each node:
WRFM_NODECPUS = compute-0-0-ib:15,compute-0-1-ib:15,compute-0-2-ib:15,compute-0-3-ib:15,compute-0-4-ib:15,compute-0-5-ib:15,compute-0-6-ib:15,compute-0-7-ib:15,compute-0-8-ib:15,compute-0-9-ib:15
Make a couple of other changes to the run_ncpus.conf file to set the type of domain decomposition and the network interface:
DECOMP = 1 HYDRA_IFACE = ib0
It may also be worth changing the NUMTILES variable from its default of 1 to a larger value for improved performance.