Arc is the primary High Performance Computing (HPC) system at The University of Texas at San Antonio (UTSA) that can be used for running data-intensive, memory-intensive, and compute-intensive jobs from a wide range of disciplines.

1. Arc is equipped with:

• 172 total compute/GPU nodes and 2 login nodes, majority of these are Intel Cascade Lake CPUs and some are AMD EPYC CPUs
• 22 GPU nodes - each containing two Intel CPUs with 20-cores each for a total of 40 cores, one V100 Nvidia GPU accelerator and 384GB of RAM
• 9 GPU nodes - each containing two Intel CPUs with 20-cores each for a total of 40 cores, two V100 Nvidia GPU accelerators and 384GB of RAM
• 3 NVIDIA DGX nodes - each having two AMD EPYC CPUs with 64-cores each for a total of 128 cores, eight Nvidia A100 80GB GPUs and 2TB of RAM
• 2 GPU nodes - each containing two Intel CPUs with 20-cores each for a total of 40 cores, four V100 Nvidia GPU accelerators and 384GB of RAM
• 2 GPU nodes - each having two AMD EPYC CPUs with 8-cores each for a total of 16 cores, one Nvidia A100 40GB GPU and 1 TB RAM
• 2 GPU nodes - each having two Intel CPUs with 24-cores each for a total of 48 cores, four Nvidia H100 80GB GPUs and 1TB of RAM
• 2 CPU nodes - each having two Intel CPUs with 24-cores each for a total of 48 cores, four Nvidia L40S 48GB GPUs and 1TB of RAM
• 2 large-memory nodes, each containing four CPUs with 20-cores each for a total of 80 cores, and each including 1.5TB of RAM
• 1 large-memory node, equipped with two AMD EPYC CPUs with 8-cores each for total of 16 cores and 2 TB of RAM
• 6 nodes equipped with two AMD EPYC CPUs with 8-cores each for a total of 16 cores and 1 TB of RAM

• 100Gb/s Infiniband connectivity

• Two Lustre filesytems: /home and /work, where /home has 110 TBs capacity and /work has 1.1 PB of capacity

• A cumulative total of 250TB of local scratch (approximately 1.5 TB of /scratch space on most compute/GPU nodes)

• Multiple partitions (or queues) having different characteristics and constraints:
  • bigmem2: 1 node
  • gpu1a100: 2 nodes
  • bigmem: 2 nodes
  • compute1: 65 nodes
  • compute2: 27 nodes
  • compute3: 6 nodes
  • gpu1v100: 22 nodes
  • gpu2v100: 9 nodes
  • gpu4v100: 2 nodes
  • two privately owned partitions consisting of 24 nodes
  • one privately owned partition equipped with three Nvidia DGX servers with 8x A100 80 GB GPUs
  • one privately owned partition equipped with two Dell R760XA servers with 4x L40S GPUs
  • two privately owned partitions each equipped with Dell XE8640 servers with 4x H100 GPUs

---

2. Arc is accessible over SSH using two-factor authentication with DUO. Hostname for Arc is arc.utsa.edu and the SSH port number is 22. In order to utilize DUO, you must register online at passphrase.utsa.edu .

---

3. Arc Fair-Use Policies

• Running Jobs
  • Compute nodes are not shared among multiple users. Instead, when a user grabs a compute node, they will be the only user allowed to access it. This is being implemented for security reasons, as well as performance reasons. If multiple users are sharing the same node, performance can be negatively impacted due to resource contention. While we will no longer be scheduling jobs from different users on the same node, users are encouraged to take advantage of tools such as GNU parallel to co-schedule their multiple independent tasks on the compute nodes allocated to them. Please see Section 10 of the user-guide for further details on running multiple tasks concurrently on one or more nodes from a single Slurm job.
  • Each user will be limited to 10 active jobs at a given point in time and will be limited to running these jobs on a maximum of 20 compute nodes. As each compute node is dual-socket, and has a 20-core processors on each socket, a total of 800 cores could be potentially used by a job at a given point in time.
  • Each job will be limited to a run-time of no more than 72 hours. Users are encouraged to consider implementing checkpointing-restart capabilities in their home-grown applications. The research computing support group will be happy to provide guidance on implementing checkpointing-restart mechanism in the users' code. Some third-party software, like the FLASH astrophysics code, already have in-built capabilities to checkpoint-restart. Such capabilities can be enabled by setting the required environment variables. The users are encouraged to review the documentation of their software to confirm whether or not the checkpoint-restart functionality is available in the software of their choice. Section 16 of this user-guide has further information on using checkpointing and restart.
  • Exceptions : If you require access to nodes for a longer period of time, or need access to more nodes than what are allowed by default, please submit a service request ticket with an exemption request. We will need a brief description of the activity for your request, along with the number of cores and nodes required, and the time duration for which you are requesting the exemption. Also, we request to explore the options for checkpointing the code before submitting the ticket for service request at the following URL: https://support.utsa.edu/myportal
• Data Storage (Disk Usage)
  • Work Directory (/work/abc123) – as detailed in our Wiki, this directory is where you should place any input/output files as well as logs for your running jobs. This directory is NOT backed up and is not intended for long-term storage.
  • Work Directory Data Retention – all files in the Work directory that have not been accessed in the last 30 days will be likely candidates for deletion.
  • Home Directory (/home/abc123) – this directory is backed up but should only be used for installing and compiling code. Storage of datasets is permitted here, but there will be a hard quota limit of 100GB in place.
  • Vault Directory (vault/research/abc123) - each user on Arc is provided 1TB of archival storage located in /vault. This storage space is accesible from Arc, as well as Windows or Mac computers. This data is backed up and the backups are replicated to UT Arlington for an extra layer of protection. If additional storage space is needed on the "vault" system, please submit a service request at the following URL: https://support.utsa.edu/myportal
• GPU Resources Utilization
  • Monitoring: We employ enhanced monitoring systems to check for GPU usage on all jobs allocated to general use GPU nodes.
  • Termination: Jobs detected to be idle or not making use of the allocated GPUs will be terminated after a grace period of 1 hour.
  • Notification: After termination, users will receive a warning email to either adjust their job accordingly or utilize one of the many CPU queues.

---

4. Requesting an Account on Arc

---

5. Prerequisite: Arc has a Linux operating system and hence, basic knowledge of Linux is required for working efficiently on Arc in command-line mode.

---

6. Onboarding Arc

• Logging Into ARC and Accessing a Compute Node
• Running Applications Interactively and Submitting Batch Jobs
• Managing Batch Jobs
• Modules for Managing User Environment on Arc
• Migrating Data from Arc to the Isilon archival storage
• General Instructions for File Transfer

---

7. Running C, C++, Fortran, Python, and R applications in Serial Mode

• Both batch and interactive modes of running serial applications is covered
• Code and scripts used in the examples shown in the document are available from this GitHub repository

---

8. Running Parallel Programs

• Code and scripts used in the examples shown in the document are available from this GitHub repository
• OpenMP, MPI, and CUDA examples are covered in this document
• C, C++ and Fortran are the base languages used

---

9. Running Multiple Copies of Executables Concurrently from the Same Job

• Running multiple executables concurrently from the same job is covered
• Using GNU Parallel for running parameter-sweep applications is covered

---

10. Additional Python and R Usage Information

• Python VMs in Anaconda
• Saving and Importing Python Environments
• Saving the Python Virtual Environment in Work Directory
• Saving and Importing R Environments: Saving the R libraries in Work Directory
• R troubleshooting - Fixing a corrupted library directory using Bioconductor

---

11. Using Some of the Popular Software Packages that are Installed System-Wide

• AI/Machine Learning Software
  • Using Tensorflow on Arc
  • Using Tensorflow with Multiple GPUs
  • Installing and Using PyTorch on Arc
• Using Matlab on Arc
• Using Abaqus on Arc
• Using Namd On Arc

---

12. Using Containers (Singularity and Docker) on Arc

---

13. Visualization Using Paraview on Arc

---

14. Setting Java Environment for Applications with Java Dependencies

---

15. Application Checkpointing and Restart on Arc

---

16. Technical Support

• For technical support, you can submit a support request for Arc at the following link: https://support.utsa.edu/myportal. Instructions for submitting support requests can be found here.
• The Research Computing Support Group is available between 8:00 AM to 5:00 PM on all business days to assist with the service requests.
• Our time-to-response on new tickets is 4 business hours, and the time-to-resolution varies depending upon the complexity of the issue.
  • Please open a new ticket for every new topic
  • Once a ticket is closed you are welcome to reopen it if the exact topic that was addressed in the ticket appears to be still unresolved
• For after-hours emergency support, please contact Tech Cafe at 210-458-5555.

---

17. Training and Workshops

References

1. "User Environment Tracking and Problem Detection with XALT," K. Agrawal, M. R. Fahey, R. McLay, and D. James, In Proceedings of the First International Workshop on HPC User Support Tools, HUST '14, Nov. 2014. dx.doi.org/10.1109/HUST.2014.6.