CCR maintains two clusters; ub-hpc and faculty. Within each cluster exist partitions (queues) that users can request to have their jobs run on. Not all partitions are available to all users. Below are descriptions of the clusters and the partitions.
Compute nodes on all CCR clusters are subject to monthly downtimes. Downtime information can be found here
Examples on how to run on these clusters and partitions can be found here.
UB-HPC Compute Cluster¶
The ub-hpc cluster contains the following partitions:
- debug: Dedicated compute nodes to help users debug their workflows
- general-compute: Large pool of compute nodes for academic users
- viz: Hardware Accelerated Graphics compute nodes
- industry: Compute nodes industry partners pay to use
- scavenger: Preemptible jobs on all ub-hpc nodes
Faculty Compute Cluster¶
This cluster contains over 50 faculty owned or project specific partitions. Access to these partitions is determined by the owner and managed via allocations in ColdFront. All idle nodes in the faculty cluster are accessible to UB-HPC users in the scavenger partition. Scavenger jobs will be preemptively canceled when jobs are submitted by members of the group that owns the node.
CCR has several node types available on our resources. Each node type is meant for certain tasks. These node types are relatively common for other HPC centers. We will discuss each node type and its intended use below.
- Use for: editing scripts, moving files, small data transfers, submitting jobs
- Do not use for: Heavy computations, building software or long running processes
- 15 Minute time limit on running processes
- Many users are typically logged into these at the same time
- Connections are balanced across two physical servers (vortex1 and vortex2)
- Use for: Building Software, submitting Jobs
- Do not use for: Heavy computations
- Access these nodes by typing
ssh compilefrom a login node
- Where jobs are executed after being submitted to the slurm scheduler
- Intended for heavy computation
- When run an interactive job will be performing tasks directly on the compute nodes
- Only users with active jobs can log in to allocated nodes
- Compute nodes reserved for testing and debugging jobs
- Accessed by submitting a debug job in slurm
- 1 Hour Walltime limit on debug jobs
- Compute nodes with Hardware Accelerated Graphics for Interactive Desktops that require OpenGL/CUDA
- Accessed through Open On Demand
Data Transfer Nodes¶
- Data Transfer Nodes (DTNs) are nodes which provide Globus data transfer services at CCR