Some Experience with WRF EMS vs.

You know something useful, tips&tricks, have some unofficial documentation...? Please share your knowledge with us in this section of EMS forum. * * * Please DON'T open topics with questions in this section! * * *
Post Reply
Posts: 8
Joined: Wed Nov 18, 2009 12:21 am

Some Experience with WRF EMS vs.

Post by godelsrevenge »

There isn't much space here but I wanted to share experience with the latest version of wrfems with forum users.

Using the installation script, the installation itself went flawlessly, as it always does. Several issues continue to be nettlesome, however. I run wrfems on a small cluster of CentOS 5.4 boxes with Opteron 248 CPUs, 2 GB RAM, and 1 TB disk space in total. I have six boxes, each box has two CPUs and one of the boxes serves as a head node.

The first obstacle I faced was simply configuring a simple cluster! ssh is not an easy piece of software, not because it has inherent difficulties but because there are almost no instructions on using it. I finally resolved those issues by creating the same user, user name, group name, and password on each box, duplicating the wrfems directory in each, then proceeding with ssh. Basically ssh for passwordless operation means creating the keys, duplicating the specific keys for each box, then signing on from all hosts to all other hosts. Simple. Hmmm. Except there are few places to find that information.

The other issue concerns the time it takes to run the model. I live in the United States in the Minneapolis, Minnesota metropolitan area. If I create two domains, with one domain comprising a portion of the metropolitan area and the parent comprising the state of Minnesota and then run wrfems with 1-2-1 smoothing for a 48 hour simulation, the model requires nearly three hours to run. If I use a somewhat larger parent domain and run a simulation with all three nests, the model requires over 24 hours to complete.

You can expect that additionally complex simulations require large amounts of time. My best guess is that the model now requires approximately 40% to 50% longer to finish. Perhaps there are some changes I could make, so I'm open to suggestions. If you are running a simulation over a large area of the globe, be prepared for long completion times. Note I am running a limited geographic area of North America. Others are probably interested in much greater geographic extents.

Michael Allen
Plymouth, Minnesota, USA
Posts: 14
Joined: Fri Aug 21, 2009 7:21 pm

Re: Some Experience with vs.

Post by Zoyx »

I saw in EMS.cshrc that you can use rsh instead of ssh. Haven't tried it yet... you may want to give it a shot. Besides being easier to config, may get rid of some overhead associated with encryption.

The line is :

setenv COMM ssh //just switch to rsh, I guess.
Post Reply