System Requirements (Processing & Imaging)
Architecture
Typical HPC architecture
- One or more headnode(s) accessed by end users to run the Insight Graphical User Interface (GUI)
- A cluster controller hosting the job scheduler and, optionally, a local license server
- One or more compute node(s) to run jobs
- High-performance storage shared between the headnode(s) and the compute node(s)
Single-Node Cluster (workstation / laptop)
For small-scale projects, a single workstation or laptop can play the roles of headnode, controller and compute node.
DUG Wave (FWI, RTM, LS-RTM) is unavailable on a simplified architecture.
System Requirements
Headnode(s)
The following configuration is for a single user.
Headnodes can be shared by multiple users, in which case system requirements increase correspondingly.
| CPU | Minimum: x86_64, 2.3GHz 8 core Recommended: x86_64, 2.3GHz 16 core |
| RAM | Minimum: 64GB Recommended: 128GB |
| GPU | Support for OpenGL 2.1+ (required for on-screen 3D visualisation). NVIDIA GPUs require the "nvidia" driver (not the default "nouveau" driver) |
| Network | 10Gbps Ethernet NIC |
| OS | Linux distribution compatible with RHEL 8 or 9 (RHEL, Rocky, ...) Microsoft Windows (limited support; see below) |
| Storage | Shared HPC filesystem (Lustre, VAST, …) User home directories must be on the shared filesystem Size depends on project requirements |
Cluster controller
| CPU | Minimum: x86_64, 2.3GHz 4 core |
| RAM | Minimum: 32GB |
| GPU | Optional |
| Network | 10Gbps Ethernet NIC |
| OS | Linux distribution compatible with RHEL 8 or 9 (RHEL, Rocky, ...) |
| Storage | Shared HPC filesystem (Lustre, VAST, …) |
Compute node(s)
Contact us for expert advice on sizing your cluster according to your project needs.
| CPU | Minimum: x86_64 Recommended: support for AVX-512 instruction set (speeds up a number of jobs including migration and SRME) A large core count is more important than clock speed. |
| RAM | Minimum: 128GB |
| GPU |
Optional
GPUs speed up a number of jobs including migration, SRME, DUG Reg, DUG Deblend, Cadzow filters and DUG Wave (FWI, RTM, LS-RTM)) |
| Network | 10Gbps Ethernet NIC |
| OS | Linux distribution compatible with RHEL 8 or 9 (RHEL, Rocky, ...) |
| Storage | Shared HPC filesystem (Lustre, VAST, …) User home directories must be on the shared filesystem |
Single-Node Cluster (Workstation/Laptop)
| CPU | Minimum: x86_64 Recommended: support for AVX-512 instruction set (speeds up a number of jobs including migration and SRME) A large core count is more important than clock speed. |
| RAM | Minimum: 128GB |
|
GPU |
Support for OpenGL 2.1+ (required for on-screen 3D visualisation) NVIDIA GPUs require the "nvidia" driver (not the default "nouveau" driver) |
| Network | - |
| OS | Linux distribution compatible with RHEL 8 or 9 (RHEL, Rocky, ...) Microsoft Windows (limited support; see below) |
| Storage | Local or network filesystem |
Cluster Software Requirements
Insight depends on a Job Scheduler to submit jobs to the cluster, even on a workstation or laptop.
Supported Job Schedulers:
- Slurm
If you use a Job Scheduler system not in the list above, please contact us at [email protected].
Access Right Requirements
The installation process requires temporary administrator access rights to the headnode(s), controller and compute node(s).
Connectivity Requirements
The cluster controller host must have outbound access to the Internet for licensing purposes. Proxies are supported.
If your cluster controller cannot have access to the Internet, please contact us at [email protected].
Cloud Compatibility
Insight Processing & Imaging (P&I) has been successfully deployed and tested on the following Cloud providers:
- DUG HPC Cloud
- Oracle Cloud Infrastructure
- Amazon AWS
- Microsoft Azure
Microsoft Windows Compatibility
Insight supports limited P&I functionality on Microsoft Windows.
Generally:
- Interactive processing is available via the Graphical User Interface and export facilities.
- Jobs are not available, except for some simple workflows using the Local Job Scheduler.
Please contact us at [email protected] for details.