Skip to content

Latest commit

 

History

History
250 lines (195 loc) · 10.1 KB

README.md

File metadata and controls

250 lines (195 loc) · 10.1 KB
Raw

NOTE: This document uses Slurm-GCP v5 version of CAE blueprint. You can also use Slurm-GCP v6 version of the CAE blueprint in this folder. it would require to append "-v6" suffix at the end of blueprint name/ deployment folder.

Computer Aided Engineering (CAE) Reference Architecture

The Computer Aided Engineering (CAE) blueprint in this folder captures a reference architecture where the right cloud components are assembled to optimally cater to the requirements of computationally-intensive CAE workloads. Specifically, it is architected around Google Cloud’s VM families that provide a high memory bandwidth and a balanced memory/flop ratio, which is particularly useful for per-core licensed CAE software. The solution caters also to large CAE use cases, requiring multiple nodes that are tightly-coupled via MPI. Special high-memory shapes support even very memory-demanding workloads with up to 16GB/core. For file IO, different Google managed high performance NFS storage services are available. For very IO demanding workloads, third party parallel file systems can be integrated. The scheduling of the workloads is done by a workload manager.

Architecture

The CAE blueprint is intended to be a starting point for more tailored explorations or installations of specific CAE codes, as provided by ISVs separately.

This blueprint features a general setup suited for CAE applications on GCP including:

  • Google's H3 VMs, ideally suited for CAE workloads
  • Google's C3-highmem VM, suited for workloads with 8GB/core requirement
  • Google's Filestore NFS-based shared storage
  • Google's Chrome Remote Desktop
  • SLURM workload scheduler

Getting Started

To explore the reference architecture, you should follow the these steps:

Before you start, make sure your prerequisites and dependencies are set up: Set up Cluster Toolkit.

For deploying the CAE reference blueprint follow the Deployment Instructions.

For pointers on how to proceed with the installation of ISV software, please see section Software Installation Patterns.

Deployment Stages

This blueprint has the following deployment groups:

  • setup: Setup backbone infrastructure such as networking, file systems, & monitoring.
  • software_installation(optional): This deployment group is a stub for custom software installation on the network storage before the cluster is brought up
  • cluster: Deploys an auto-scaling cluster and remote desktop.

Having multiple deployment groups decouples the life cycle of some infrastructure. For example a) you can tear down the cluster while leaving the storage intact and b) you can build software before you deploy your cluster.

Deployment Instructions

Warning

Installing this blueprint uses the following billable components of Google Cloud:

  • Compute Engine
  • Filestore

To avoid continued billing after use closely follow the teardown instructions. To generate a cost estimate based on your projected usage, use the pricing calculator.

[!WARNING] Before attempting to execute the following instructions, it is important to consider your project's quota. The cae-slurm.yaml blueprint creates an autoscaling cluster that, when fully scaled up, can deploy up to 10 h3-standard-88 and up to 10 c3-highmem-176 VMs.

To fully scale up this cluster, the project would require quota for:

  • Compute Node Group
    • 88 CPUs * 10 VMs = 880 H3 CPUs
    • 176 CPUs * 10 VMs = 1760 C3 CPUs
  • Remote Desktop Group
    • 40 N1 CPUs
    • 5 T4 GPUs
  • Slurm Login & Controller VM
    • 24 N2 CPUs
  • Filestore
    • 2x Basic SSD
    • 1x High Scale SSD

However, this is merely an example sizing for an instance of this reference architecture. Node counts and remote desktop seats can easily be adjusted in the blueprint.

  1. Clone the repo

    git clone https://github.com/GoogleCloudPlatform/cluster-toolkit.git
cd cluster-toolkit

  2. Build the Cluster Toolkit

    make

  3. Generate the deployment folder after replacing <project> with the project id.

    ./gcluster create community/examples/cae-slurm.yaml -w --vars project_id=<project>

  4. Deploy the setup group

    Call the following gcluster command to deploy the cae-slurm blueprint.

    ./gcluster deploy cae-slurm

    The next gcluster prompt will ask you to display, apply, stop, or continue without applying the setup group. Select 'apply'.

    This group will create a network and file systems to be used by the cluster.

    [!WARNING] This gcluster command will run through 2 deployment groups (3 if you populate & activate the software_installation stage) and prompt you to apply each one. If the command is cancelled or exited by accident before finishing, it can be rerun to continue deploying the blueprint.

  5. Deploy the software_installation group (optional).

    [!NOTE] Installation processes differ between applications. Some come as a precompiled binary with all dependencies included, others may need to be built from source, while others can be deployed through package managers such as spack. This deployment group is intended to be used if the software installation process requires substantial amount of time (e.g. compilation from source). By building the software in a separate deployment group, this process can be done before the cluster is up, minimizing costs.

    [!NOTE] By default, this deployment group is disabled in the reference design. See Software Installation Patterns for more information.

    If this deployment group is used (needs to be uncomment in the blueprint first), you can return to the gcluster command which will ask you to display, apply, stop, or continue without applying the software_installation group. Select 'apply'.

  6. Deploy the cluster group

    The next gcluster prompt will ask you to display, apply, stop, or continue without applying the cluster group. Select 'apply'.

    This deployment group contains the Slurm cluster and with compute partitions and a partition for Chrome remote desktop visualization nodes.

  7. Set up Chrome Remote Desktop

    One or multiple Chrome Remote Desktop (CRD) sessions can be started dynamically through Slurm.

Teardown Instructions

Note

If you created a new project for testing of the CAE solution, the easiest way to eliminate billing is to delete the project.

When you would like to tear down the deployment, each stage must be destroyed. Since the software_installation and cluster depend on the network deployed in the setup stage, they must be destroyed first. You can use the following commands to destroy the deployment in this reverse order. You will be prompted to confirm the deletion of each stage.

./gcluster destroy cae-slurm

Warning

If you do not destroy all three deployment groups then there may be continued associated costs.

Software Installation Patterns

This section is intended to illustrate how software can be installed in the context of the CAE reference solution.

Depending on the software you want to use, different installation paths may be required.

  • Installation with binary Commercial-off-the-shelf CAE applications typically come with precompiled binaries which are provided by the ISV.

    See the tutorials for Ansys Fluent and for Siemens Simcenter STAR-CCM+ that illustrate this process.

    In general, you need to bring the binaries to your CAE cluster for which it is useful to use a Google Clouds Storage bucket, which is accessible from any machine using the gsutil command and which can be mounted in the cluster.

    As this installation process only needs to be done once and at the same time may require time, we recommend to do this installation in a separate deployment group before you bring up the cluster. The `software_installation' stage is meant to accommodate this. You can for example bring up a dedicated VM

    - id: sw-installer-vm
  source: modules/compute/vm-instance
  use: [network1, appsfs]
  settings:
    name_prefix: sw-installer
    add_deployment_name_before_prefix: true
    threads_per_core: 2
    machine_type: c2-standard-16

    where you can follow the installation steps manually. Or using the toolkit's startup-script module, the process can be automated.

    Once that is completed, the software will persist on the NFS Filestore share for as long as you do not destroy the setup stage.

  • Installation from source/with package manager For open source software, you may want to compile the software from scratch or use a package manager such as spack for the installation. This process typically takes a non-negligible amount of time (~hours). We therefore strongly suggest to use the software_installation stage for this purpose.

    Please see the HCLS Blueprint example for how the software_installation stage can be used to use the spack package manager to install all dependencies for a particular version of the software, including compiling the software or its dependencies from source.

    Please also see the OpenFOAM example for how this can be used to install the OpenFOAM software.

    Once that is completed, the software will persist on the NFS Filestore share for as long as you do not destroy the setup stage.