Skip to content

Latest commit

 

History

History
102 lines (70 loc) · 4.86 KB

aws-pcluster-paraview.md

File metadata and controls

102 lines (70 loc) · 4.86 KB

Cluster Rendering using AWS ParallelCluster and IndeX for ParaView plugin

Custom Amazon Machine Image (AMI):

  • Go to the NVIDIA IndeX offering in the AWS Marketplace and subscribe to the image.
  • Once subscribed, please get the AMI id for your region.
  • Note: AMI is associated with an Amazon EC2 Region. An instance needs to be launched in the same region as the AMI.

ParallelCluster Configuration:

  • If you are new to AWS ParallelCluster refer to this blog

  • To install AWS ParallelCluster refer to this guide

    pip3 install aws-parallelcluster==2.8.0 --upgrade --user

    Important: The custom AMI provided is tied to a specific ParallelCluster version. So make sure to install the specific ParallelCluster version, in this case 2.8.0.

  • Please use the parallel cluster configuration template to create your config file.

  • Once you customized the config file, please copy it to ~/.parallelcluster/config.

  • Note: the template configuration enables remote desktop visualization using NICE DCV. To learn more about about NICE DCV refer here

Create the cluster as:

  • To create the cluster, please run the following command:

    pcluster create <cluster-name> -c <path-to-config-file>

  • The given config template creates a cluster with 4 p3.8xlarge instances for the compute nodes (each with 4 V100 GPUs) and 1 g4dn.4xlarge instance for the master node. You can always modify the instance types in the configuration through the compute_instance_type and master_instance_type fields.

Connect to the Remote desktop (NICE DCV server) running on the master instance:

  • pcluster dcv connect <cluster-name> -k <your-key.pem>
  • This will start a Remote Desktop session and connect to the master instance using a web browser

Install ParaView:

  • Run the utility script to install ParaView with NVIDIA IndeX enabled from a shared directory: 
    cd /shared
/opt/scripts/install-paraview.sh

It's important to install ParaView to a shared directory so that compute nodes have it as well.

Cluster Rendering using ParaView:

  • Here we will use ParaView in client-server configuration.

  • Run a slurm job to start ParaView server on <nnodes> number of nodes with <gpus_per_node>:

    salloc -N<nnodes> --exclusive
srun --mpi=pmi2 -N<nnodes> -n<gpus_per_node> --gpus-per-task=1 /opt/scripts/xauth.sh /shared/ParaView-5.8.1-MPI-Linux-Python3.7-64bit/bin/pvserver --force-offscreen-rendering -display :0

This should output:

Waiting for client...
Connection URL: cs://ip-<master-rank-ip>:11111
Accepting connection(s): ip-<master-rank-ip>:11111

  • Start the ParaView client and connect it to the server session started above:

    /shared/ParaView-5.8.1-MPI-Linux-Python3.7-64bit/bin/paraview --server-url=cs://ip-<master-rank-ip>:11111

ParaView will start now connected to the pvserver cluster. If you go back to the pvserver slum job output, you should see the client is connected.

  • In ParaView Client, Click on View → Memory Inspector. This will display the cluster and rank assignments

Loading the Sample Dataset

  • Fetch the sample supernova dataset in a shared location:
cd ~/Downloads
wget https://nvindex-datasets-us-west2.s3-us-west-2.amazonaws.com/scenes/00-supernova_ncsa_small/data/Export_entropy_633x633x633_uint8_T1074.raw

  • Click on File → Open → <path-to> → Export_entropy_633x633x633_uint8_T1074.raw

    • Click OK
    • Open Data with Image Reader

  • Update the properties for this data set.

    • The data set used here is unsigned char data type
    • Confirm the Data Byte Order for your system (LittleEndian vs. BigEndian). You can use this to find out. For example, on a x86 system you would select Little Endian.
    • Data Extent is the X, Y, Z dimension of the dataset (its specified in the name of file). For this dataset it would be [0, 632] for X, Y and Z dimensions.

  • Change colouring from SolidColor to ImageFile.

  • Change representation to use the NVIDIA Index renderer: Click on the Outline dropdown and select NVIDIA IndeX.

  • At this point you should see a colured cube. Change the data range (via colormap or Rescale to Data Range button) to to [25, 255].

  • At this point you should see the features of the dataset. Feel free to use the colormap to highlight different features.

  • Here's a sample rendering screenshot:

  • The dataset shown here is a time step in a core-collapse supernovae simulation. Credits