Getting started

How can I get a CCAST account?

Please apply for a CCAST account by filling out this form.

How do I login to CCAST?

You can login to CCAST via

• Secure Shell (SSH) at prime.ccast.ndsu.edu or thunder.ccast.ndsu.edu
• Web browser at https://ondemand.ccast.ndsu.edu

See here for more details.

I forgot my password. Can you reset it for me?

No, CCAST is a service provider only and does not have the ability to reset passwords. Follow the instructions in this article to reset your password. If that doesn’t work, contact the NDSU IT Help Desk.

How much does CCAST cost to use?

The basic level of services is free of charge to NDSU faculty, staff, and students as well as certain external collaborators (upon approval of CCAST’s Executive Director). You won’t receive any surprise bill for services from CCAST. Faculty members may purchase dedicated computing capacity for their research program if they need, but many users are able to get by with the resources that are freely available.

What is the difference between “Thunder” and “Thunder Prime”? Which one should I use?

“Thunder” and “Thunder Prime” are CCAST’s two High-Performance Computing (HPC) clusters.

Thunder is an Intel-based cluster, procured in 2013 and expanded several times up until 2020. Since 2020, no new hardware or software is being installed in Thunder, but it is still functional and performant for many types of computational research.

Thunder Prime is an AMD-based cluster, procured in 2020 and expanded in 2022. It is NDSU’s flagship HPC cluster, containing roughly 9,000 CPU cores and 50 Nvidia Ampere GPUs. All new hardware and software installations are targeted for Thunder Prime.

Both clusters share the same 2 PB GPFS file system, allowing for data to be used in computations on either cluster.

Whether to use Thunder or Thunder Prime depends on the specific requirements of the user’s workload, including considerations such as CPU architecture, required number of cores, memory capacity, and GPU utilization. For more details, please see the Thunder and Thunder Prime system description pages.

What is the Linux command for doing X?

A basic list of Linux commands is available here.

A cheat sheet specifically for the PBS Pro batch scheduler can be found here.

In addition, most commands on the system have a manual page, which can be accessed from the command line using man <command>. For instance, the manual page for the ls command can be accessed using man ls. The full online manual for Linux is also available at die.net.

Do you provide any training or resources for people who are new to HPC?

Yes, CCAST provides training workshops each semester covering the fundamentals of HPC, from beginner to intermediate. Keep an eye out for announcements on the CCAST user mailing list and other academic department mailing lists.

We also have a section on “Research Computing and Support” in the NDSU IT Knowledge Base (KB). See here for a listing of our articles and tutorials.

How can I get help?

The best way to get help is to email ndsu.ccast.support@ndsu.edu. If you are facing a specific issue, please provide a detailed description of the problem, including screenshots, error messages, and/or text output if available. If the issue is related to a failed or problematic batch job or interactive session, please also provide the job/session ID.

Data storage and transfer

How much data can I store on CCAST?

CCAST has the following default limits for data storage:

• Home (/mmfs1/home/$USER): 200GB
• Scratch (/mmfs1/scratch/$USER): 20TB
• Projects (/mmfs1/projects/<project_owner>): 1TB

When you first log into the CCAST system, your current disk usage is displayed. You can also check your disk usage by logging into Open OnDemand at selecting “Utilities > Disk Usage” from the navbar.

How do I transfer files between my local machine and CCAST?

If connecting to CCAST using a text terminal (e.g., PuTTY, PowerShell, Bash) you can use the Secure Copy (scp) command to transfer files. If connecting via web browser (Open OnDemand), you can use the “Files” app to view and transfer files. There are also a number of graphical file transfer clients, such as WinSCP or FileZilla. For larger transfers, you can use the Globus file transfer service.

For more information on transferring files, please see the Transferring Files section of the CCAST User Guide.

I used up all my storage space. Can I have more?

No, home and scratch quotas are fixed for all users and cannot be increased.

However, if you are a project owner on CCAST (faculty and certain staff), you may request an increase to your project space by completing the appropriate quota increase request form.

Can I share data with other users of CCAST?

Yes, data on CCAST can be shared within project directories among members of the same project group. You can either add a user to a project group you own, or request to be added to someone else’s project directory. Requests for project group changes should be submitted to ndsu.ccast.support@ndsu.edu. Ideally, these are initiated by the project group owner, but they can also be initiated by another person with the project owner copied for approval.

Running jobs

How do I submit a job on CCAST?

Jobs are defined by “job scripts” and are submitted using the qsub command. A successfully submitted job will be assigned a job ID, which can be used to check the status of the job. Examples of job scripts for popular software packages can be found in /mmfs1/projects/ccastest/examples/.

Users can also submit interactive jobs using the interact command. This launches an interactive terminal session on a compute node. Different options can be specified for interact to adjust the number of CPU cores, amount of memory, and other attributes. See interact -h for a list of valid options.

For more information on submitting jobs, please see the Running Jobs section of the CCAST User Guide.

How can I see what resources are currently available?

The freenodes command shows which nodes are available and how many free resources they each have. There are several optional arguments that can be provided to freenodes that provide more or less information. Run freenodes -h to show the help menu.

What is a queue? Which one should I submit my job to?

A queue refers to a collection of nodes that have similar characteristics, or accept similar types of jobs. The following queues can accommodate what most users need:

• default: Primary queue for CPU jobs. Many nodes belong to the default queue.
• gpus: Contains nodes that have GPUs installed. If you need a GPU, you have to submit to this queue.
• interactive: Nodes that can run interactive sessions, either in the terminal or via Open OnDemand.
• preemptible: Contains every node in the cluster, but jobs submitted to this queue can be interrupted at any time, without warning.
• condo queues: Contain nodes owned by individual research groups. Not available for use by regular users, except via the preemptible queue.

Please see the Running Jobs section of the CCAST User Guide for more information.

What is the maximum amount of resources I can request for a job?

Maximums for each resource type depend on the queue you are submitting to, but for most queues, the following maximum limits apply:

• CPU cores: 128
• Memory: 1000GB
• GPUs: 4
• Walltime: 168:00:00 (168 hours, or 7 days)
• Nodes: as many as are available in the queue

Keep in mind, however, that just because you request a certain number of resources does not mean your code will be able to use all those resources. For example, if you are running a code that is single-threaded (i.e., can only use one CPU core), it will not run any faster whether you request one core or 100 cores. Similarly, if your code is not built to use GPUs, requesting a GPU will not accelerate your code.

If your code can use multiple cores/GPUs, or even multiple nodes (e.g., MPI codes), then you can request multiple of these types of resources. However, the more resources you request, the longer it will take the scheduler to find those resources, meaning your job may wait in the queue for a longer period of time.

How do I specify a project group for a job submission?

Each job requires a project group to be defined for submission, which can be specified by setting the following option in your job script:

#PBS -W group_list=<project_group>

To see which project groups you are a member of, you can run the myprojects command in the terminal, or click on “My Project Groups” in the “Utilities” menu on Open OnDemand.

How can I check the status of my job?

You can check on the status of a job using the qstat command. When run by itself (no additional arguments), qstat lists all jobs currently in the queue.

To restrict the output to only your own jobs, run qstat -u $USER. To get output for only a single job, run qstat <jobID>.

Can I run multiple jobs at once?

Yes, users can run multiple jobs simultaneously. The maximum is 250 concurrent jobs per user.

Can I request a specific CPU or GPU architecture for my job?

Yes, you can add the following options to your job’s select statement:

• For CPU architecture: plist=<cpu_model>
• For GPU architecture: glist=<gpu_model>

To find which CPU and GPU architectures are available, run freenodes -cg and look at the options available in the “CPU” and “GPU” columns, respectively.

How can I troubleshoot a failed job?

If your job fails, first look in the job’s working directory (usually the directory where the job was submitted) for any output or error files. These files often contain messages or progress reports that may indicate what the job was doing when it failed, along with any errors it may have encountered.

If the issue is not clear based on the output files, or if your job didn’t produce any output files, you can contact ndsu.ccast.support@ndsu.edu for assistance. Be sure to include the job ID, a copy of the job script, and a description of the problem you are facing.

How do I cancel one of my jobs?

Use the qdel (queue delete) command and specify the ID of the job you want to delete:

qdel <jobID>

My job has been waiting in the queue for a long time. Why isn’t it running?

You may have requested a very large amount of resources and the scheduler has to wait for other jobs to finish before enough resources will be available to run your job. Or you may have submitted a resource request that is impossible to fulfill on any combination of nodes (e.g., requesting 200 CPU cores per node, when the largest node only has 128 CPU cores).

Run qstat -fx <job ID> to get detailed information about your job, and look for the “comment” field at the bottom. If you see an error that says “Can never run”, you need to double check your resource request and modify it to something more reasonable.

If you’re not sure what the problem is, contact ndsu.ccast.support@ndsu.edu for help.

How can I make my job run faster?

First, make sure you are differentiating between queue time. If your job is taking a long time to finish because it is waiting in the queue a long time before running, see the previous FAQ for tips on reducing your wait time.

If your job is taking a long time to run, and wait time is not the issue, look at the documentation for your software to see if it can take advantage of parallel processing. Many software packages have additional options to specify a number of CPU cores to use. This is the easiest way to speed up a process. Your software may also be able to take advantage of multiple nodes, using something like MPI. Again, read the documentation to see the recommended implementation.

If neither of the above are options with your software, you may still be able to speed up your code if you can break the problem into smaller pieces, and process each of those pieces in a separate job.

Software

How can I check which software packages are available on CCAST?

Software on CCAST is made available through a “modules” system. To see a list of available software modules, run:

module avail

If you want to search for a particular piece of software, you can supply a keyword. For example:

module avail python

How do I load a software module?

If your software is available on CCAST via a module, you can load it with the following command:

module load <software/version>

The version portion is optional, but recommended. If you do not specify a version, the system will load the latest version of the software, which may change if a newer version is installed.

I don’t see the software I need in the modules list. Can you install it for me?

If a package is generally useful to a large portion of the user community, or is needed by several different research groups, CCAST staff may be able to install it for you. Please email ndsu.ccast.support@ndsu.edu with your software installation request, including information such as the name of the software, a link to the software’s website, and a brief description of your use case.

Can I install my own software on CCAST?

Yes, many software packages can be installed in your home directory or your principal investigator’s project directory without administrative privileges (i.e. without sudo) using the available compilers and libraries. It is your responsibility to make sure that for any software you install yourself, you are complying with the applicable license (whether open-source or commercial) and that use of the software is consistent with NDSU IT acceptable use policies.

Common error messages (and how to fix them)

“Disk quota exceeded”

This error indicates one of your storage directories is full and cannot store any more data. You can check your disk usage by logging into Open OnDemand and selecting “Disk Usage” from the “Utilities” menu.

If you are over your quota in one of your personal directories (i.e. home or scratch), you will need to either delete files from the directory to free up space, or migrate some of your files to a different location, such as a project directory.

If you are a project owner and you are exceeding your project directory quota, you may be eligible for a quota increase. Request an increase by filling out this form.

“Permission denied” (when installing software)

This means you are trying to install something in a location you do not have permission to modify, such as a global system directory. For many software packages, you can choose an alternative install directory that you own, such as your home or project directory. Read the documentation for the particular software you are installing to determine the relevant option(s) needed to change the install location.

“qsub: Bad GID for job execution”

This error indicates that you are submitting a job with a project group that you are not a member of. To see which project groups you are a member of, you can run the myprojects command in the terminal, or click on “My Project Groups” in the “Utilities” menu on Open OnDemand. Then make sure you specify one of these groups in your job script:

#PBS -W group_list=<project_group>

“qsub: Job rejected due to undefined group list”

This error means you neglected to specify a project group for job submission. See the previous question for instructions on how to properly specify a project group in your job script.