Tutorials at SXScon 2024

This repo contains workshop material like Jupyter notebooks and input files used for the SXS workshop at ICERM. It should be cloned on Oscar, the cluster at Brown which we will be using. To access Oscar, see the instructions below.

The documentation for Oscar can be found here. These docs are quite good, so if there's anything missing from this README about how to do something on Oscar, you can refer here. Or you can also ask any organizer/TA for help.

1. Connecting to Oscar

You should already have an account on the cluster. At the beginning of the workshop you should have received an email from Brown/ICERM with details about a Brown account. This should have your username and password. These will be the username and password you use for everything below.

Duo 2FA

Brown requires 2FA for all its accounts. To setup 2FA, go to your brown account, log in with your credentials, and follow the instructions there.

Wifi

There are two networks you can use. The guest wifi requires no authentication. To use the Brown network, you'll have to log in with your Brown username and password. Access to the Brown network will be required if you want to use ssh keys to access Oscar, or if you want to use VSCode to access Oscar.

Logging in to Oscar with a terminal

To access oscar, simply do

ssh [email protected]

You'll be prompted for you password, and then for 2FA.

SSH Keys (Optional)

Using ssh keys is not required for the workshop. Though it can make accessing Oscar easier so you don't have to type your password in every time.

⚠️ You can only use ssh keys to access the sshcampus.ccv.brown.edu domain. And you can only access the sshcampus.ccv.brown.edu domain from the Brown network. Ssh keys will not work for the ssh.ccv.brown.edu domain or on the guest wifi.

If you don't already have an ssh key, the GitHub docs are very easy to follow.

We need to copy our key to Oscar

ssh-copy-id -i ~/.ssh/id_ed25519 [email protected]

You should be prompted for a password and 2FA. After this, you should be able to access Oscar without entering a password

ssh [email protected]

Logging in to Oscar with VSCode (Optional)

You cannot log in to Oscar with VSCode through either of the above domains (sshcampus.ccv.brown.edu or ssh.ccv.brown.edu). You also cannot use VSCode to access Oscar on the guest wifi. You have to be on the Brown network.

You will need to add the following to your ~/.ssh/config file

Host oscar-vscode
  HostName vscode1
  User USERNAME
  ProxyJump oscar-jump

Host oscar-jump
  HostName poodcit4.services.brown.edu
  User USERNAME

replacing the username with your own. Then with the remote ssh extension , find the oscar-vscode host, and connect with that. If you set up ssh keys, then you won't be prompted for a password. If you didn't set them up, you'll have to enter your password and 2FA.

2. Clone this repo

Once you have access to Oscar, clone this repo

git clone https://github.com/sxs-collaboration/sxscon-2024-tutorials.git $HOME/sxscon-2024-tutorials

3. Getting an interactive compute node

Oscar offers a convenient command for getting an interactive compute node

For Jupyter and paraview, use

interact -f cascade -n 2 -t 01:00:00

For building SpECTRE use

interact -f cascade -n 8 -m 16G -t 01:00:00

The -n means number of cores, -t time limit. You can optionally also specify the amount of memory with -m 10G. This will give you 10GB of memory.

For building SpECTRE, you probably want to use 8 cores and 16G of memory. For the Jupyter and ParaView servers, you can get away with 2 cores and the default amount of memory.

4. Getting SpECTRE set up for the first time

1. Download SpECTRE:

   git clone https://github.com/sxs-collaboration/spectre.git $HOME/spectre

2. Configure your environment on Oscar:

   echo 'export SPECTRE_HOME=$HOME/spectre' >> ~/.bashrc
   echo 'export SPECTRE_BUILD=$HOME/spectre/build' >> ~/.bashrc
   echo 'export PATH=$SPECTRE_BUILD/bin:$PATH' >> ~/.bashrc
   echo 'source $SPECTRE_HOME/support/Environments/oscar.sh' >> ~/.bashrc
   source ~/.bashrc

   (You'll want to do the next steps on a compute node).

3. Load the SpECTRE modules:

   spectre_load_modules

   To make sure things loaded correctly run module list. You should see modules like spectre-deps/oscar-2024-07 and spec-exporter/spec-2024-07 loaded.

4. Configure the SpECTRE build directory:

   mkdir -p $SPECTRE_BUILD
   cd $SPECTRE_BUILD
   spectre_run_cmake

5. Now you can compile the code (on a compute node, 8 cores, 16G memory). Compile the command-line interface (cli), the binary black hole executables (bbh), and several additional executables that we'll use in the tutorials:

   cd $SPECTRE_BUILD
   make -j8 cli
   make -j8 bbh
   make -j8 EvolveBurgers EvolveNewtonianEuler1D EvolveScalarWave1D

   Things should compile quite quickly because we've set up a cache on Oscar. You can check the cache hit rate with ccache -s.

Getting set up again after disconnecting

Just run:

spectre_load_modules

Then you can compile and run the code, open Jupyter notebooks, etc.

Troubleshooting

If you run into issues, please ask a TA. Unfortunately with shared file systems on clusters issues do arise. An easy thing to try when you run into issues is to wipe the build directory and recompile:

1. Delete the build directory (make sure you don't have any files in the build directory that you want to keep):

   rm -rf $SPECTRE_BUILD

2. Go to step 4. above to reconfigure the build directory.

5. Connecting Jupyter Lab to Oscar

See OpenOnDeman for an alternative method.

You have to run Jupyter notebooks/lab on a compute node since they can sometimes have a lot of computation.

We offer a convenience bash script to help setup the Jupyter server. On the compute node

cd $HOME/sxscon-2024-tutorials
./jupyter_setup.sh $SPECTRE_BUILD

After this, you'll need to open a new ssh connection in a separate terminal. This script prints instructions on how to do this, and then how to open the server in a local browser. You can also open the URL in VSCode if it is connected to Oscar.

6. Connecting Paraview to Oscar

See OpenOnDeman for an alternative method. (Note you'll) have to install your own paraview 5.11.2 on the interactive node since the system installed one doesn't have GUI capabilities

Unfortunately paraview is very strict when trying to do remote rendering about the local and remote versions of the code being identical. This means you'll need to download a specific version of paraview for this workshop, even if you already have a different version installed.

You specifically need version 5.11.2 (https://www.paraview.org/download/).

If you are on linux, you can download it with

wget https://www.paraview.org/paraview-downloads/download.php?submit=Download&version=v5.11&type=binary&os=Linux&downloadFile=ParaView-5.11.2-MPI-Linux-Python3.9-x86_64.tar.gz
tar -xzf Paraview-5.11.2*

On macOS, choose the correct arch for your laptop (arm vs x86). The new M(1/2/3) macs need arm. Older macs use x86.

Similar to running a jupyter server, paraview needs to be run on a compute node as well.

We also offer a convenience bash script to help set this up

./paraview_setup.sh

If the port you were assigned is busy, you can rerun the script or pass a specific port to the script. The output of the script will also tell you what to run on your local machine to connect your local paraview to the server (it'll also be a new ssh connection). There are different commands to run depending whether you are on linux or macOS.

7. OpenOnDemand

You can also access oscar using OpenOnDemand . This may be useful if you're having issues accessing the cluster. It runs on a compute node and opens a GUI locally so you don't have to deal with port forwarding. Just be sure to grab enough resources if you need to build SpECTRE.

Log in to here with your Brown credentials. Then click the 'Desktop' button. There will be a dropdown for how many resources to use. Choose whichever you like. Wait a few seconds, then click the 'Launch' button. If your internet connection is good, you can decrease the compression and increase the image quality. Then you should be dropped into a nice GUI! You can open a terminal and do everything from there without worrying about port forwarding.

Note that for both Jupyter and Paraview, you don't need to run the _setup.sh scripts anymore. You can just run

$SPECTRE_BUILD/bin/python-spectre -m jupyterlab

or wherever you installed paraview (wget https://www.paraview.org/paraview-downloads/download.php?submit=Download&version=v5.11&type=binary&os=Linux&downloadFile=ParaView-5.11.2-MPI-Linux-Python3.9-x86_64.tar.gz and untar it)

./bin/paraview