Migrating to Hive

How Hive is different

Hive supports the HI-Performance Personnel Onboarding portal (Hippo), hippo.ucdavis.edu. Hippo offers user account and PI group creation through a web-based interface. PIs can request to become a sponsor, which allows their users to request access to PI owned resources. Please note that Hive does not support the Genome Center Computing Portal. Changes made within the Computing Portal will not have any effect after migration. 

Hive uses the Quobyte clustered file system for enhanced performance, reliability, and ease of administration. Quobyte offers a single namespace for data, so PIs can get expanded storage without migration to a different file storage system or adding a new file share. Please note that Quobyte only supports MPI-IO with MPICH or OpenMPI version 5 and above. Quobyte does not support encryption, so files in directories that are migrated might take up more space on Hive.

Hive allocates each user a 20 GB home directory storage space for code and configuration. Data should be stored in the PI group storage.

Hive supports the use of SSH with campus passphrases. SSH keys are supported, but not required. They can be uploaded/updated through Hippo and will automatically deploy.

Hive supports Open OnDemand, a web-based graphical interface with a file browser: JuypterLab, RStudio, VS Code, and a Linux desktop. OnDemand allows users to run jobs on compute nodes through their web browser. Open OnDemand is currently in alpha testing mode, and availability is not guaranteed.

Hive supports high and low partitions. High partitions are reserved for labs that have purchased resources; low partitions are available to all users pending resource availability. Hive does not support a medium partition. The medium partition on older clusters was possible due to a bug in the job scheduler code, which has since been patched.

Please note: 

To ensure a more secure and efficient environment, we are transitioning from direct SSH access to individual compute nodes to a centralized access system via Slurm’s srun --pty functionality. This change brings several benefits:

  1. Simplified Resource Management: Slurm provides a streamlined way to access and allocate resources, making workflows more predictable and efficient.
  2. Consistent User Experience: The Slurm interface ensures a unified access method, simplifying the process for users and administrators alike.
  3. Reduced resource contention: since resource allocation happens through Slurm, users will not be able to overwhelm or crash a system.
  4. Improved Security: By eliminating direct server SSH access, we reduce the attack surface and strengthen overall system security.

This change supports our commitment to providing a secure, reliable, and user-friendly environment.

Maintenance / Downtime Schedule

Hive undergoes scheduled maintenance twice yearly in coordination with the Campus Data Center generator test. The generator test is scheduled for approximately one week in Spring and December after finals. Services will be down and unavailable for the duration of maintenance.

Migrate Your Home Directory

Users must create an account on Hive using the Hippo user portal, and are responsible for migrating content from their home directory to Hive. Users who do not have a UC Davis affiliation will need to have a temporary affiliate (TAF) account sponsored and approved by their PI. Users can transfer files via SSH or rsync. Globus is available, but must be requested for each PI share.

Migrating Storage

If you've bought resources on the Genome Center's Barbera (LSSC0) cluster, the data must be migrated to Hive's Quobyte file system. This is a multi-step process which entails:

  1. All migrating users who don't currently have an account on Hive must create a new account via the user portal, Hippo. Users who do not have a UC Davis affiliation will need to have a temporary affiliate (TAF) account sponsored by their PI and have it approved.
  2. Pre-staging data. HPC staff will migrate data from Barbera to Hive. Depending on the amount of data, this can take hours, days, or weeks. Typical network speeds for migration have taken an hour per terabyte.
  3. Quiescing group storage and jobs for a final migration. HPC staff will work with each lab to arrange for downtime to copy over any remaining files that have changed from pre-stage to final migration. Again, this can take several hours or days, depending on the data size.
  4. Changing file ownership. Hive is a new cluster, and group and user IDs will differ. Converting the data from Barbera ownership to Hive ownership requires a manual process. 
  5. Validation process: PIs must work with HPC staff to validate the correct data and permissions.

Migrating Login Nodes and Partitions

Migrating to Hive is easy. Administrators will provision resources on Hive equal to supported log-in nodes. Please note that all users must create an account with the Hippo portal. Users will need to migrate data from their home directories to Hive. Please remember that home directories on Hive have a 20G limit. 

Log-in nodes will integrate into Hive as standard High partition nodes.

Requesting Software

Hive uses Spack and the Cern VM File System to compile and serve software modules. The HPC team utilizes Spack to build software against a standard set of dependencies and makes the resulting modules available through CernVM-FS. CernVM-FS is a multi-tiered system that uses an upstream proxy server to make copies of the file system available to each HPC cluster that HPC@UCD maintains.

Users are encouraged to install and maintain their own software by using Conda in their lab's share or the user's home directory. A central conda install is provided by typing: module load conda.

Users can request software to be installed as a centrally installed module within CernVM-FS if it meets the following criteria:

  • Software must benefit the majority of users on the HPC cluster. Single-user software will not be considered.
  • Software must not have licensing terms that require HPC to accept responsibility for use or installation.
  • Selected software must be robust enough to run without error, as HPC will not fix immature software.

To request a software install, please read the software install policy, which contains the link to the software install request form. Since HPC is inundated with software install requests, it can take 2-3 weeks to attend to them. Please be patient.