DINE
The Durham Interconnected Novel Environment (DINE) is a test production cluster aimed at testing different HPC, networking and storage technologies. It is now in the second generation.
To date, it has hosted BlueField-1, Rockport and BlueField-2 technologies.
Installed in 2019, DINE was the first UK AMD Rome based production cluster.
Specifications
The DINE cluster originally had 24 nodes, each equipped with a 200G BlueField-2 Data Processing Unit from NVIDIA. 18 are currently in use, in the bluefield1 Slurm partition (b101-b119, without b115).
DPUs offload networking and storage workloads from the main CPU of the node, separating infrastructure processing from job processing.
Nodes |
RAM per node |
CPU(s) |
Network Technology |
|---|---|---|---|
18 |
512GB |
2x AMD EPYC 7302 16-Core |
Infiniband |
Usage
If you do not already have an account on COSMA, please follow the instructions here, then request to join the project: do009.
The DINE cluster is part of the bluefield1 SLURM partition, which accepts the do008, do009 and durham projects. Jobs should be submitted to the queue using #SBATCH -p bluefield1. The same nodes are also in the dine partition, for project dp128. See queues.
There is no direct ssh access to DINE nodes.
Known issues / notes
The bluefield nodes are automatically powered down when not in use, so when you submit your job you may see the status as CF (configuring) while your allocated nodes are booting.