A better method is to run the impact_bmad container here, which has a conda env with lume impact setup inside it and then navigate to your lume impact notebook and run it once in jupyterlab in the container.
Examples here
Additional information on these examples: Running Impact-T on s3df
Useful links:
Current release info
Installing Non-MPI Lume-Impact
follow steps listed inThis method uses tunneling to connect to jupyter lab. To use jupyter lab through ondemand additional steps need to be done.
Prerequisites:
ssh to the iana node with tunneling enable:
Ssh into s3df login node with port forwarding (5555 can be changed to any preferred port):
Code Block ssh -L 5555:localhost:5555 <SLAC UNIX USERNAME>@s3dflogin.slac.stanford.edu
Connect to Iana compute cluster :
Code Block ssh -L 5555:localhost:5555 iana
(First time only) Install conda via miniconda:
This will install conda to the home directory
Code Block wget https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh -O /tmp/Miniconda3-latest-Linux-x86_64.sh
Code Block bash /tmp/Miniconda3-latest-Linux-x86_64.sh -p /sdf/home/c/<UNIX_USERNAME>/conda/
Create a new Conda environment
Code Block conda create -n <NEW_ENV_NAME> -c conda-forge python=3.8 dask distributed ipykernel numpy scipy deap h5py pint matplotlib mpi4py
Install Lume-Impact for the conda environment used to activate jupyterlab as shown below:
Code Block |
---|
conda activate <NEW_ENV_NAME> |
lume-impact
from the conda-forge
channel can be achieved by adding conda-forge
to your channels with:Code Block |
---|
conda config --add channels conda-forge conda config --set channel_priority strict |
Once the conda-forge
channel has been enabled, lume-impact
can be installed with:
Code Block |
---|
conda install lume-impact |
Install other packages (optional):
Code Block |
---|
conda config --add channels conda-forge conda install pydantic=1.10 conda install xopt=1.4.1 |
It is possible to list all of the versions of lume-impact
available on your platform with:
Code Block |
---|
conda search lume-impact --channel conda-forge |
Run a jupyterlab server with the impact conda env activated:
- ** do only first time ** install jupyter lab if not installed:
Code Block |
---|
conda install jupyterlab |
- Run jupyter and click on the link to open in your browser:
Code Block |
---|
jupyter lab --no-browser --port 5555 |
Impact-T Executables
Here impact is the conda env you are using.
Impact-T is available through conda-forge and can be installed via:
conda
create
-n
impact
source
activate impact
or conda activate impact
For non-MPI
conda
install
-c
conda-forge
impact-t
After
these
steps,
the
IMPACT-T
executable
ImpactTexe
or
ImpactTexe-mpi
,
respectively,
will
be
in
your
environment
variable
and
is
thus
ready
to
use
like
any
regular
command-line
command.
If the Impact-T executables are still not found, then you should go to the conda env folder, find the executables and manually add them to your path.
Visit https://github.com/impact-lbl/IMPACT-T for these and further instructions, including those to build from source.
MPI Lume-Impact
To use run Lume-Impact in parallel one can use mpi. These steps follow this notebook.
- install the mpi version of impact-t instead of the regular impact-t when creating the conda env (nb: this is the same env with xopt, jupyterlab, lume-impact etc)
Code Block conda install -c conda-forge impact-t=*=mpi_openmpi*
3. configure the settings to point to where your impact-t executable is. It should be in the conda env files. Here eric used an conda env called xopt
Code Block 'command': '/sdf/home/c/cropp/conda/envs/xopt/bin/ImpactTexe', 'command_mpi': '/sdf/home/c/cropp/conda/envs/xopt/bin/ImpactTexe-mpi',
More information here: impact-T