Connect to DASH Cluster
- Connect your computer to a Duke Health network
- Connect on-premise through a Duke Health wired or wireless connection or connect remotely using VPN (
vpn.duhs.duke.edu
).
- Connect on-premise through a Duke Health wired or wireless connection or connect remotely using VPN (
SSH to the cluster login node
ssh <NetID>@dash.duhs.duke.edu (or @somhpc.duhs.duke.edu)
Successful SSH show show you a welcome message:
$ ssh <NetID>@dash.duhs.duke.edu @dash.duhs.duke.edu's password: xxxx Last login: Mon May 16 03:58:32 2022 from 10.38.0.70 #################################################################### # Welcome to DASH (DHTS Azure School of Medicine HPC) Shared Cluster #################################################################### .... ....
A quick glance of the cluster nodes status using sinfo to find a node for running your job
EXAMPLE: $ sinfo PARTITION AVAIL TIMELIMIT NODES STATE NODELIST execute* up infinite 8 idle~ dash1-execute-[3-12] execute* up infinite 2 idle dash1-execute-[1-2] gpu up infinite 4 idle~ dash1-gpu [1-3,5] gpu up infinite 1 mix dash1-gpu-4 highmem up infinite 2 idle~ dash1-highmem-[1-3]
To request an interactive session, always use nodes dash1-execute-[1-2]. For sbatch jobs you can use any node in execute partition. Highmem partition is reserved for high compute and high memory workload and incurs higher costs to run jobs. Read more about the DASH cluster specifications here.
NOTE: DASH does not have a dedicated interactive partition like HARDAC which limits jobs to 24 hours. So, CLOSE your interactive sessions when you are done with them to prevent unnecessary charges.
Descriptions of STATE:
- mix: node is in use
- idle: node is not in use, but is ready to run a job
- idle~: node is not in use, but the VM is not ready to run a job (average time to wake up a VM is 8 minutes)