Using compute.cla

Connect

Connect to lts.cla.umn.edu using the NX client (see z.umn.edu/ltsconnect).

Decide Session Type

  • Use an interactive session when you need to see what is happening - e.g. using a graphical application
  • Use a batch session when you have a script to run
  • Use a batch session when you need more power or longer runtimes

Interactive Session

 

Note that these instructions have changed and SSHing to compute.cla.umn.edu is no longer needed.  You may enter qsub commands from lts.cla.umn.edu.

Open a Terminal

Enter the command qsub -IX to start an interactive session using the default resources (8GB of memory, 24 hour walltime, and 2 cores on one node). Please note that at the end of the "walltime" your session is terminated so resources are available for others. Adjust your walltime as needed (see below table for limits).

The example below starts an interactive session, loads the module for MATLAB, and then runs MATLAB. (Most of our software is now provided via modules.  Please see using modules for information on how to load and use software via modules.)

 goldy@lts:~> qsub -IX
 c5:~> module load matlab
 c5:~> matlab
 
In the above command, the "-I" tells it you want an interactive session and the "-X" tells it you want graphical programs forwarded to your screen. compute.cla will find a node with resources and you’ll be at the command prompt of that node (nodes are numbered C1 - C8). From here you can run software just as you would on any of our Linux servers.
 

If you find you need a more powerful interactive session you can ask qsub to give you more resources by following the example below. This command will give you 16GB of memory, 48 hours, and 8 cores on one node.

goldy@lts:~> qsub -IX -l mem=16GB,walltime=48:00:00,nodes=1:ppn=8

Batch

The batch queue is a routing queue that routes jobs to the short, long, highmem, or multinode queue based on the resources requested for the job. This simplifies job  submission and frees you from having to determine which queue is the correct one to use for your job.

Using compute.cla for batch jobs is similar to interactive in command but you drop the -IX, add the queue (-q batch), and add the name of the script you want to run.

In the example below, we have a script called analysis.sh. To run it, enter this command:

 compute:~> qsub -q batch analysis.sh

The following example shows the command for a batch job requesting specific resources.

compute:~> qsub -q batch -l mem=16GB,walltime=48:00:00,nodes=1:ppn=8 analysis.sh

Using a PBS script

While you can submit your job to the cluster using a shell script as above, specifying the resource list on the command line each time can be tedious and doesn't lend itself well to repeatability.  The recommended way to submit a job involves creating a PBS job submission script which contains the setup information and resources needed as well as the commands to be executed. The PBS script is then submitted to the cluster using the qsub command. More information on the general format of a PBS job submission script as well as some examples can be found on our page for creating a PBS script.

Launching Applications

In general, the latest applications are best used via modules. See our page on using modules for more information.
If the app you need isn't in modules, software is located in /pkg/software_depot/app and /pkg/software_depot/bliss (neuroimaging specific apps)

Options

 
 -l walltime=hh:mm:ss Requests the amount of time needed for the job. Default is 24 hours.  At the end of the walltime, the session terminates.
 -l nodes=n:ppn=p Requests number of nodes and processors per node. The range of ppn values depends on the hardware you are running on.  If this option is not used the default is 1 node and 4 cores
 -l mem=amount Request the total amount of memory needed across all nodes. e.g. mem=4GB
 -q queuename Requests a particular queue.  Without this, it defaults to the Interactive queue
 

Queues and Resource Limits

The table below shows the default and maximum compute resources for each queue. The batch queue is a routing queue that routes jobs to the short, long, highmem, or multinode queue based on the resources requested for the job. Note that submitting jobs directly to the short, long, highmem, or multinode queues isn't allowed as all jobs intended for those queues are routed to them via the batch queue. The interactive and batch queues are available to everyone but access to the highmem, GPU and multinode queues is restricted. If you find you're needing more resources than what the short and long queues provide, let us know and we can add you the highmem or multinode queues as needed. Access to the GPU queue is also restricted so If you would like to use the GPU, please let us know so that we can provide you access to that queue.

Queue Name

Access Control List

Direct Submission Allowed?

Interactive?

Default

Maximum

Walltime

RAM (GB)

Procs

Walltime

RAM (GB)

Procs

interactive

None

Yes

Yes

48:00:00

8

2

168:00:00

64

10

batch

None

Yes

No

 --

short None No No 48:00:00 8 2 95:59:59 128 20
long None No No 96:00:00 8 2 336:00:00 64 10

highmem

Yes

No

No

24:00:00

32

2

95:59:59

250

40

multinode

Yes

No

No

04:00:00

 8

2

168:00:00

750

120

gpu Yes Yes Yes  
gpu-prod Yes No Yes 08:00:00 32 2 48:00:00 96 20
gpu-dev Yes No Yes 04:00:00 8 2 04:00:00 8 2