Uploaded image for project: 'SAFe Program'
  1. SAFe Program
  2. SP-4278

On-Demand Scratch in Azimuth using OpenStack Manila

Change Owns to Parent OfsSet start and due date...
    XporterXMLWordPrintable

Details

    • SRCnet
    • Hide

      Several UKSRC demonstrator cases have a need for a large amount of local scratch storage for the duration of their processing. This type of storage is also called out within the SRCNet roadmap.

      OpenStack Manila is able to provide on demand filesystems that are compatible with the "hard" multi-tenancy between OpenStack projects (and Azimuth tenancies). Commercial storage solutions such as PureStorage and Vast are known to have similar capabilities for providing isolated storage e.g. via Kubernetes CSI driver, although not always currently supported by OpenStack Manila.

      The advantage of the "hard" multi-tenancy is that user are able to have root access within Azimuth platforms. Moreover, there should be no way to "break out" of your tenancy to see other people's storage. This enables many development use cases and helps reduce the level of trust given to users of these platforms. However, it does make sharing storage more difficult compared to tweaking POSIX file permissions.

      While we focus on an end-to-end user experience via Azimuth, this gives us a better understanding of how this could be integrated using other tooling, such as the IVO Execution broker or CANFAR.

      Show
      Several UKSRC demonstrator cases have a need for a large amount of local scratch storage for the duration of their processing. This type of storage is also called out within the SRCNet roadmap. OpenStack Manila is able to provide on demand filesystems that are compatible with the "hard" multi-tenancy between OpenStack projects (and Azimuth tenancies). Commercial storage solutions such as PureStorage and Vast are known to have similar capabilities for providing isolated storage e.g. via Kubernetes CSI driver, although not always currently supported by OpenStack Manila. The advantage of the "hard" multi-tenancy is that user are able to have root access within Azimuth platforms. Moreover, there should be no way to "break out" of your tenancy to see other people's storage. This enables many development use cases and helps reduce the level of trust given to users of these platforms. However, it does make sharing storage more difficult compared to tweaking POSIX file permissions. While we focus on an end-to-end user experience via Azimuth, this gives us a better understanding of how this could be integrated using other tooling, such as the IVO Execution broker or CANFAR.
    • Hide

      AC1: Azimuth user scan create scratch storage and attach it to multiple Azimuth platforms, including: Workstation and Slurm.

      AC2: Azimuth user can delete Azimuth storage, with an appropriate warning when the storage is likely still being used by other platforms.

      AC3: Get feedback on how well this approach is working with the UKSRC demonstrator cases, and share that with the wider SRCNet team via the HPC and Cloud COP.

      AC4: (stretch goal) consider registering shares as a Rucio RSE to help "schedule" copy in of data, potentially via FTS.

      Show
      AC1: Azimuth user scan create scratch storage and attach it to multiple Azimuth platforms, including: Workstation and Slurm. AC2: Azimuth user can delete Azimuth storage, with an appropriate warning when the storage is likely still being used by other platforms. AC3: Get feedback on how well this approach is working with the UKSRC demonstrator cases, and share that with the wider SRCNet team via the HPC and Cloud COP. AC4: (stretch goal) consider registering shares as a Rucio RSE to help "schedule" copy in of data, potentially via FTS.

    • PI23 - UNCOVERED

    • SRCNet0.x operations-and-infrastructure team_DAAC

    Description

      UKSRC have been supporting various workflows using pathfinder/precursor data. Some recent LOFAR sky survey work has show then need for easy access to large scratch areas for the duration of processing.

      Typically, this scratch storage is generally needed first to copy in data, then attach to a large compute cluster to process the data, then attached to something else to copy out the results to persistent storage.

      OpenStack Manila allows on demand creation of file systems, that can be isolated within each tenancy, while allowing users to retail root access, if required. Where the user has root level access, we cannot use POSIX permissions on a shared filesystem to isolate data, because the user would be able to bypass these restrictions.

      We need to compare this approach with VOSpace / CANFAR offerings.

      The specific scenario we would like to test is Azimuth Slurm and Workstation appliances where:

      • Datalake of read-only source data is available (stored remotely)
      • User creates (read-write) scratch storage space of the required size (e.g. 20TB)
      • Workstation used to copy in required data sets
      • Slurm used to process data sets
      • Worksation used to copy out results into Datalake
      • Scratch filesystem deleted, and freed up for another research group

      Attachments

        Structure

          Activity

            People

              b.mort Mort, Ben
              D.Watson Watson, Duncan
              Votes:
              0 Vote for this issue
              Watchers:
              1 Start watching this issue

              Feature Progress

                Story Point Burn-up: (0%)

                Feature Estimate: 0.0

                IssuesStory Points
                To Do00.0
                In Progress   00.0
                Complete00.0
                Total00.0

                Dates

                  Created:
                  Updated:

                  Structure Helper Panel