Draft: UCSC Shared HPC Policies and Guidelines

End User License Agreement (EULA)

These are draft policies and guidelines which have been presented to the Research Computing and Data Infrastructure Committee for comment. Community comments or inquiries can be directed to Research IT. These policies and guidelines are in effect, though enforcement may vary while the policies are in draft form.

By using UCSC shared high-performance computing (HPC) resources you (the “user”) agree to the following.

Accounts

  1. UCSC shared HPC has two account types: open access accounts and sponsored access accounts.
    1. Open access accounts are provisioned automatically and have access to Hummingbird Open Access Cluster (HOAC). Any user with valid UCSC credentials and access to the campus VPN may log into HOAC. Open access accounts are bound by the same EULA as other types of accounts, except where superseded by sponsored access.
    2. Sponsored access accounts require a faculty sponsor. Sponsors are responsible for properly vetting users working under their supervision and for the behavior of these users. Sponsors must have a current Export Control Attestation on file. Sponsored access accounts have privileged access to advanced HPC nodes (i.e. nodes connected to parallel file system at 200Gbps) as well as access to legacy HOAC nodes.
  2. All users agree that systems administrators or sponsors for their account are entitled to regulate, limit, suspend, or terminate their access.
  3. Access to sponsored accounts that are inactive for more than one year may have their access revoked. HOAC accounts older than one year may be deleted.
  4. Users agree to acknowledge the use of UCSC resources in publications. The following wording is recommended: “This research used resources of the shared high performance computing facility at the University of California, Santa Cruz.”
  5. Both account types use campus Gold authentication. Users agree to protect their access credentials in accordance with University network and security policies.
  6. Users agree to comply with applicable UC Santa Cruz policies, contracts, and legal obligations, including intellectual property rights and confidentiality agreements.
  7. Groups and shared folders for HOAC require faculty sponsorship
  8. Users agree that violations of the policies in this EULA may result in loss of access to UCSC HPC resources, including termination of accounts.
  9. Accounts which are associated with users who no longer have active UCSC accounts (e.g. matriculated students or separated faculty) will be purged on a quarterly basis. Users are responsible for making copies of their home folders and any associated data before their UCSC accounts are disabled.

Usage

  1. Shared HPC resources are to be used primarily for research. They can be used for instruction only with prior approval.
  2. Users are encouraged to acquaint themselves with standard HPC practice by attending HPC training, HPC Open Office Hours, and engaging with the HPC user community through the UCSC HPC (Hummingbird-cluster) Slack Channel. For more information on HPC training, please visit https://hummingbird.ucsc.edu
  3. UC Santa Cruz shared HPC resources should not be used to process, store, or transmit confidential or regulated information as defined in University of California Office of the President (UCOP) policy IS-3. In addition, please note:
    1. UCSC shared HPC resources do not meet the standards of NIST CUI, DoD NISPOM, HIPAA, EAR, or ITAR, and should not be used to process, store, or transmit information governed by these regulations. 
    2. Users that access data subject to other agencies’ policies (e.g. NIH Genomic Data Sharing Policy) agree to comply with the terms and conditions for use of such data.
  4. UCSC shared HPC resources use the module system. Only HPC systems administrators are authorized to install software as a module. Users may request any valid software be installed by the systems administrators, but final determination on the suitability and practicality of such installations shall be determined by staff. Users are encouraged to use this system to ensure the reliability and reproducibility of their computations. Software installed as a module is available globally.
  5. All applications and installations must be run on compute nodes and should either be submitted to a job queue using the Slurm command sbatch. Applications or installs running on any login node may be subject to immediate termination without notice. Repeated or intentional abuse of this policy may result in account suspension or termination.
  6. Licensed software may only be installed on shared HPC resources if the license permits its use for research on multiple computers across a network. Users are responsible for ensuring that any software they install or request to be installed on shared HPC resources abides by license agreements. For guidance on licensing software and vendor assessment, please refer to UCOP’s Appendix for Data Security.
  7. Because these resources are shared among all users, users should gauge their needs accurately and avoid oversubscribing in their requests for compute resources. UC Santa Cruz HPC administrators monitor usage and may notify users that they are subject to job suspension or termination that fail to use resources judiciously.

Investing in our shared HPC resources

  1. Investors are defined as principal investigators who have made purchases of compute, storage or network infrastructure that has been integrated into the shared HPC resources.
    1. Investors have priority access to the nodes they purchase and exclusive access to stand-alone storage they purchase. 
    2. Jobs running in the shared backfill partition, that spans all nodes and which allows the use of idle resources by all users, are subject to preemption. 
    3. General users are responsible for check-pointing their backfill jobs to be able to restart efficiently in the event of preemption. All users are encouraged to check-point their work.
  2. Compute, storage and networking infrastructure is subject to architectural review to assess its suitability to purpose, complexity and other features that harmonize with hardware standards set forth by the Research Computing and Data Infrastructure Committee.
  3. Purchases compute nodes are encouraged to aggregate their purchases with other investors, so that orders can be filled economically using bulk discounts where possible.
  4. Investor nodes whose warranty has expired will remain in service at the owner’s discretion, so long as they are maintained in a functional state by the owner (in coordination with research IT staff), and so long as there are sufficient resources in the data center facility to support them. The node’s owner will be responsible for providing any needed funds for replacement of parts that are out-of-warranty. Staff will coordinate their purchase and installation. Disposition of decommissioned nodes will be determined by the investor, but it is recommended that they be recycled responsibly.

Storage

  1. Each user is provided a home folder of 1TB at no cost. In addition, is principal investigators are provided a lab folder calculated based on the number of active lab HPC users (n), typically n*1 TB by request.  Exceptions to this policy for labs which require additional storage are at the discretion of the Director of Research IT.
  2. Data on home and lab folders is not backed up. While we make efforts to snapshot these directories in case of catastrophic failure, it is not recommended to store primary data on any shared HPC resource covered by the EULA. Researchers are expected to maintain fail-safe copies on other platforms.
  3. Temporary scratch storage is available to all users for their jobs. Scratch directories are subject to automatic deletion, without notice, when their time limit of two weeks expires or when the storage partition requires housekeeping.

Maintenance

  1. The use of shared HPC resources and services is at the user’s own risk. While we  will use best efforts to maintain a functional research computing resource, we make no guarantee that resources or services will be available at all times, that they will suit every purpose, or that data will never be lost or corrupted.
  2. Shared HPC resources may be periodically unavailable for scheduled maintenance. Notice of these events will be provided. Routine maintenance will be timed to minimize impact on research activities, but may occur at any time.
  3. UCSC shared HPC resources may be unavailable without notice, and running jobs may be subject to termination, if hardware, software, or other issues necessitate immediate maintenance to prevent interruption or loss of service.
  4. Users shall comply with requests by UCSC shared HPC staff regarding job submission and resource usage that are needed to prevent interruption or loss of service.

UC Santa Cruz Research Computing