High Performance Computing
Three-dimensional computer tomography images can consist of 2000³ or even more voxels. GeoDict allows to compute flow, thermal, electrical or mechanical properties on such large datasets. In these computations, multiple floating point values per grid cell must be computed and stored. As 2000³ already means 8 billion grid cells, such computations need strong computational resources. These simulations require:
- a large amount of memory (RAM),
- a high number of cores to be computed in a short amount of time and
- a large hard drive to store the simulation results.
These requirements can be met by large shared memory machines, or computer clusters. Often, such computational resources are not available locally, and there is a need for on-demand availability of computer resources in the cloud.
Shared Memory Machines
Large shared memory machines are single computers with a large amount of RAM, many CPU cores and one or more large hard drives. These machines have the advantage that they are relatively cheap to purchase and easy to maintain. They allow to perform and visualize simulations on very large structures. Shared memory machines are the best option for most of the application cases. Recommendations about the hardware configurations for different use cases can be found at:
https://www.geodict.com/service-support/technical-support/system-requirements.html
The operating system on these machines may be Windows or Linux. A GeoDict installation on such a shared memory machine can be used by several users. You can use a remote desktop connection to connect to this computer, or alternative use GeoDict’s built-in job queue system to submit compute jobs from a local desktop to this computer.
Be aware that multiple licenses are required if several users want to use GeoDict at the same time or if a single user wants to use GeoDict on several computers at the same time (local desktop and remote server). .
|
Computer Clusters
A computer cluster typically consists of compute nodes with the same hardware configuration which are communicating with each other over a very fast interconnection (e.g. InfiniBand).
A floating license is required to use GeoDict on such clusters. Be aware that multiple licenses are required if several users want to use GeoDict at the same time or if a single user wants to submit several GeoDict computations to the clusters job system at the same time.
GeoDict uses MPI (Message Passing Interface) to exchange data between different compute nodes (distributed memory). For this, a supported MPI version has to be installed on the cluster. Not all of Geodicts solvers are able to use multiple compute nodes, see the Cluster Computing topic for details.
|
Topics of this section
|
- Parallelization Options
Description of the Parallelization settings dialog, that lets you select the number of parallel processes used in many GeoDict commands.
- MPI Configuration
Learn how to install MPI on a Linux computer and how to select the appropriate MPI version.
- Remote Desktop
To enable 3D rendering on a remote Linux computer, install TurboVNC and Virtual GL.
- Cluster Computing
Use the PBS or SLURM job system to submit jobs on a distributed memory Linux cluster.
|
