TOMAS setup guide
This page describes how to acquire the latest source code, data, and libraries required to build and run GEOS-Chem with TOMAS aerosol microphysics on the ace-net glooscap cluster.
Overview
The latest public release of GEOS-Chem with TOMAS does not include many of the recent developments in aerosol science. It also cannot take advantage of parallel computing technologies. However, the 'bleeding edge' code has many recent developments in GEOS-Chem/TOMAS that are not included in the public release, including parallel computing.
Getting Set Up
Compiler
GEOS-Chem works best with the Intel Ifort Fortran compiler - v11.1 There is an instance of the compiler installed on glooscap, which you can load by doing
module load intel/11.1.073
Alternatively, I have installed ifort version 11.1.080. To use this version, add the following to your .bashrc
export LD_LIBRARY_PATH="/home/sfarina/geos-chem-libraries-intel11/lib" export PATH="/home/sfarina/geos-chem-libraries-intel11/Compiler/11.1/080/bin/intel64:/home/sfarina/opt/bin:$PATH" export LD_LIBRARY_PATH="/usr/local/gnu/lib64:/usr/local/gnu/lib:/home/sfarina/geos-chem-libraries-intel11/lib:/home/sfarina/geos-chem-libraries-intel11/Compiler/11.1/080/lib/intel64/:/home/sfarina/geos-chem-libraries-intel11/Compiler/11.1/080/idb/lib/intel64" export INTEL_LICENSE_FILE="/home/sfarina/geos-chem-libraries-intel11/software/intel/Compiler/11.1/080/Licenses" source /home/sfarina/geos-chem-libraries-intel11/Compiler/11.1/080/bin/ifortvars.sh intel64 export FC="ifort"
Code
You can grab the absolute latest code from my source directory on glooscap:
cp -r /home/sfarina/source/GC_Bleeding_Edge/ ~
or, (safer) you can grab my latest "snapshot"
cp /home/sfarina/source/GC_BE_snapshot-latest.tgz .
Libraries
geos-chem-libraries-intel11 is a bundle of software required to build and run the latest version of GEOS-Chem. Included in this package:
- NetCDF - Network Common Data Format libraries - required to read and write certain datasets
- HDF5 - Hierarchical Data Format - required to read and write certain datasets
- other dependencies - required for netcdf and hdf5
Please follow the directions for Installing libraries for GEOS-Chem wiki before proceeding. You will need to install the netCDF-4.2 libraries.
Environment
After installing the libraries, your .bashrc should include a similar section to the following
ROOT_LIBRARY_DIR="/home/sfarina/geos-chem-libraries-intel11" GC_BIN=$ROOT_LIBRARY_DIR/bin GC_INCLUDE=$ROOT_LIBRARY_DIR/include GC_LIB=$ROOT_LIBRARY_DIR/lib export GC_BIN export GC_INCLUDE export GC_LIB
For reasons I can't recall, I use a specific version of the ifort11 compiler that is not installed by default on glooscap. If you cannot successfully compile and run geoschem with tomas using the intel ifort 11 installed on glooscap (using $ module load intel/11.1.073), adding the following to your .bashrc should allow you to use the version that I have installed with Sajeev's help:
export FC="ifort" export LD_LIBRARY_PATH="/home/sfarina/geos-chem-libraries-intel11/lib" export PATH="/home/sfarina/geos-chem-libraries-intel11/Compiler/11.1/080/bin/intel64:/home/sfarina/opt/bin:$PATH" export LD_LIBRARY_PATH="/usr/local/gnu/lib64:/usr/local/gnu/lib:/home/sfarina/geos-chem-libraries-intel11/lib:/home/sfarina/geos-chem-libraries-intel11/Compiler/11.1/080/lib/intel64/:/home/sfarina/geos-chem-libraries-intel11/Compiler/11.1/080/idb/lib/intel64" export INTEL_LICENSE_FILE="/home/sfarina/geos-chem-libraries-intel11/software/intel/Compiler/11.1/080/Licenses" source /home/sfarina/geos-chem-libraries-intel11/Compiler/11.1/080/bin/ifortvars.sh intel64
Once the compiler and libraries are installed in ~/geos-chem-libraries-intel11
source ~/.bashrc ifort --version
If ifort returns
ifort (IFORT) 11.1 20101201
you should be all set to start compiling
Data
To set up the necessary data for GEOS-Chem, simply
cd ~ ln -s /home/sfarina/data .
This will allow you to link to my data directory, which is mostly a collection of links to the data at /home/rmartin/group/ctm/ with some changes due to recent GC development. DO NOT copy this directory, as it is many many many gigabytes, and is probably beyond your disk quota on glooscap.
Building GEOS-Chem/TOMAS
Make
Glooscap allows you to use multicore interactive shells to do heavy processing. I invoke a 16 core shell to build geoschem. put this in your .bashrc
alias pshell16="qrsh -V -cwd -l h_rt=08:00:00 -l h_vmem=2.0G -l h_stack=12.5G -N IA_16 -pe openmp 16 bash" alias pshell8="qrsh -V -cwd -l h_rt=08:00:00 -l h_vmem=2.0G -l h_stack=12.5G -N IA_8 -pe openmp 8 bash"
Then you can do
cd YOUR_CODE_DIR/GC_Bleeding_Edge/GeosCore pshell16 make -j16 tomas40
This will build GEOS-Chem with 40 bin TOMAS using 16 processors at a time. As an added bonus, this will not choke up the rest of the users on glooscap.
The available target names are:
tomas <--TOMAS 30 tomas12 tomas15 tomas40
Alternatively, you can use the following to define a tomas version when compiling:
make TOMAS=yes geos make TOMAS40=yes geos etc.
Important!
When changing tomas versions, always always always do
make realclean
Running GEOS-Chem with TOMAS
Run Directories
There are run directories for each of the tomas versions at:
/net/samqfs/pierce/sfarina/standard_run_directories/
Copy the tarballs (named 40.tgz, 30.tgz, etc.) to a standard location. You can then do
tar zxvf YOUR_STANDARD_LOCATION/40.tgz
to extract the appropriate run directory to your current working directory. The folder will be named run.TOMASXX, where XX is 12, 15,30, or 40 depending on the version you would like to run.
Once you have the appropriate version of geostomas compiled and your run directory extracted, copy the executable to your run directory.
input.geos
The input.geos file is where most of the runtime options for geoschem are configured. There are currently no TOMAS specific entries in the input.geos file, save for diagnostic output quantities. Please see the Users' Guide for more information.
Submitting Jobs to the Parallel Queue
In each folder is a file called parallel.sh. Below is a description of some of the parameters:
#!/bin/bash # $ -S /bin/bash ./etc/profile #$ -o job_output #$ -l h_rt=100:00:00 #wall clock time requested from grid engine. Lower request times will have higher priority in the queue #$ -l h_vmem=2.0G #vmem requested from grid engine. 2.0 is sufficient for all versions at 4x5 and TOMAS15 at 2x2.5 on 16 cores #$ -l h_stack=12.5G #stack memory requested from grid engine #$ -N RUN_NAM #a name for your run #$ -pe openmp 16 #number of cores you are requesting from grid engine #$ -cwd #inherit properties from your current shell export OMP_NUM_THREADS=16 #number of openMP threads export KMP_STACKSIZE=500000000 #stacksize memory limit for each thread ulimit -t unlimited # cputime ulimit -f unlimited # filesize ulimit -c unlimited # coredumpsize ulimit -m unlimited # memoryuse ulimit -l unlimited # memorylocked cd YOUR_RUN DIRECTORY ./geostomas > log
You'll need to edit it slightly (run name and working directory), then run:
qsub parallel.sh
You can check on the status in the queue with
qstat
You can watch the logfile output of your simulation with
tail -f log
With some minimal editing, you can find some summary information from your runs using the script here
/net/samqfs/pierce/sfarina/testruns/informed/hourstat.sh
A Note about Speed
Choosing the appropriate version of tomas for your needs includes consideration of time and resources. Using 16 processors on glooscap at 4x5 resolution, the model time : real time ratio is roughly as follows:
version | speedup 40 bin - 64 30 bin - 82 15 bin - 144 12 bin - 170
Developing
Writing for GEOS_Chem is pretty straightforward. Please try to follow the style guide as much as possible. Most of TOMAS is contained within tomas_mod.F90, and you should be able to find what you need with a little work and a few invocations of grep. If you can't find what you need, ask.
Version Control
Git! You should definitely use git to track your changes. To use git on glooscap:
module load git
Branching and Commits
Once you have your source code directory, make a separate branch for yourself before making any changes. This will simplify trading and tracking updates/advances/bugfixes.
git checkout -b MY_NEW_BRANCH vi fictional_example_mod.F90 git status git add fictional_example_mod.F90 git commit
Patching
If I make some new changes to my branch of code, you will need to do a patch and merge. My current branch in git is called tomasmerge. If I provide you with update.patch, this should do the trick:
git checkout tomasmerge git apply update.patch git checkout MY_BRANCH git merge tomasmerge
Reference
There are many useful resources for git on the web. Here are some I found useful:
Debugging
There are two major ways of debugging: inserting massive amounts of print statements, or using a debugger. Both are useful.
ifort comes with a debugger similar to gdb: iidb.
geos-chem-libraries-intel11/Compiler/11.1/080/bin/intel64/iidb
In order to use it, you must compile geostomas as follows
make realclean make DEBUG=yes tomas
Apart from the debugger and normal print statements, TOMAS has a very useful builtin called DEBUGPRINT, that prints the values of the TOMAS size bins in a big table.
Post Processing
Now that you've successfully run the model, there are a few more hurdles to inspect your data.
Installing IDL
Copy the IDL / gamap scripts from my home directory.
cp -r ~sfarina/IDL ~
Edit the following as needed, and add it to your .bashrc
IDL_STARTUP="/home/sfarina/IDL/idl_startup/idl_startup.pro" IDL_DIR="/usr/local/itt/idl/idl80/" IDL_PATH="$IDL_DIR:/home/sfarina/IDL" module load idl/8.0
Processing
GEOS-Chem currently outputs all data in the form of a binary punch file (.bpch). These files must be handled using IDL. The process is outlined below:
Copy
Copy the relevant files to your postprocessing directory for a given run
ctm.bpch diaginfo.dat tracerinfo.dat proc_one.pro averageCNCCN_XX.py <-- XX is TOMAS version plotCNCCN.py
Split
Use the script Bpch_Sep_Sal interactively from within the IDL environment to ctm.bpch into separate months For example, to extract august, 2005 from ctm.bpch
idl > Bpch_Sep_Sal,'ctm.bpch','ctm.08.bpch',Tau0=nymd2tau(20050801) > exit
Create netcdf output
Using the IDL script proc_one.pro, we extract information from the monthly .bpch files and save it to the standard netCDF Edit proc_one.pro to use the correct infile/outfiles Execute proc_one from your shell:
idl proc_one.pro
Counting CN and CCN
Run averageCNCCN_XX.py, where XX is the model version For example, to bin and average the August results from TOMAS15:
./averageCNCCN_15.py 08
Plotting the Results
Edit your directory name to be of the format YYY_run.TOMASXX, where YYY is a run number, and XX is the TOMAS version. plotCNCCN.py will automatically detect the model version and customize map names. To plot the surface and zonal average concentrations of CN3, CN10, CN40, and CN80 for august:
./plotCNCCN.py 08
Once you have completed this process, you will have a zonal and surface level map of CN3, CN10, CN40 and CN80 predicted by the model.
NCview
You can also use ncview on the file ctm.nc to view individual species concentrations or nucleation rates.
ncview ctm.nc ncview ctm_nuc.nc
Other Advice / Issues
- If you have followed these instructions and geoschem crashes without any output, try (un)commenting the "welcome to geoschem" and the following call flush lines from main.F. This is a hard to track bug related to ongoing development of grid independent geoschem.
- I use the GNU Bourne Again SHell (bash). I suggest you do the same. The csh is fine, but I have written all of my scripts using bash. Your life will probably be easier if you use bash.
- If you are trying to run geoschem outside of a qrsh (grid engine) environment (i.e. on the head node), you will need to add ulimit -S -s unlimited to your .bashrc
- It is a good idea to TAKE NOTES on the details of your simulations.
- Making a backup of your code and any important files is a good idea. Making two backups is a better idea.
- if you have any questions or you are running into trouble, please ask either myself, Sajeev, or Jeff for help. I am usually able to respond to emails within a day, and am willing to use gchat or skype if need be.
--Salvatore Farina 17:28, 25 July 2013 (EDT)