Its possible to start up each of the components of the model system "by hand" and see them operate without the overhead of the workbench. (You don't need Java or Globus.)
There are three possible configurations:
flow2d by itself using fake topography and precip. precipsend -> coupler -> flow2d MM5 -> mm5coupler -> coupler -> flow2d
Start by downloading the tar files for each program. There is a README for each program describing how to compile. They use an autoconf script and Makefile similar to those written by Nick for the field library.
NOTE: you may need the GNU version of make to build the programs.
NOTE TO SGI USERS: there are some problems compiling flow2d.c with gcc under SGIs. autoconf will use gcc if it finds it. To force it to use the native c compiler, set the environment variable CC to point to the native c compiler: "setenv CC /opt/MIPSpro/bin/cc". Then do a "make distclean", "./configure" and "make".
Assuming you've already built an executable following the README instructions: edit flow2d.c and undefine the cpp constants COUP and RTERR. type 'make flow2d' Try a simple case with 100 meter resolution using the default egg-carton topography and running for 6 hours with a 2 min timestep: >flow2d -dx 100 -T 21600 -dt 120 -v A file called runoff.v5d will be created.
Assume you run each component on a different host. host1 for precipsend, host2 for the coupler and host3 for flow2d. Things will be more clear if you use a seperate window/xterm for each host/program.
Untar and build each model on its host. (Be sure COUP and RTERR are defined at the top of flow2d.c) Issue the following commands on each host: host1> presend host2> coupler -inhost host1 -fhost host3 host3> flow2d -host host2
You'll see the precipsender send precip for the first 24 hours of simulated time then exit. flow2d will continue for another 24 hours. Flow2d will output a vis5d file showing the computed runoff.
NOTE: This configuration has not been fully tested. All the pieces have been hooked together and run but the coupler has not been properly positioned under the precip generated by the model.
Again assume you run each component on a different host: host1 for MM5, host2 for mm5coupler, host3 for the coupler and host4 for flow2d.
Build each component including the socket enabled version of MM5.
Starting the parallel MM5 will vary from system to system. The example scripts provided here work on the Origin 2000 at NCSA.
The sample namelist will run the Chicago test case on a single grid for 24 hours, outputting precip every timestep (two minutes). This configuration was used because it can finish in a demo-length amount of time. The 2-grid simulations take longer.
Edit the sample namelist and replace HOSTNAME with the the full name of the host mm5coupler is running on. If using the sample runscript, copy the namelist back to the mass store.
Now assuming all the MM5 data and executables are in place on the mass store and all the component models are ready to go, you can start the full coupled system with the following commands (Again assuming the NCSA Origin is being used):
host1> bsub < mpi.mm5.floodsc98 host2> mm5coupler host3> coupler -inhost host2 -fhost host4 host4> flow2d -host host3
As in the precipsend case, MM5 will output data for 24 simulated hours and then stop. The runoff model will continue for another 24 simulated hours.