MSI Terminology

The following words and phrases are used throughout the MSI website and unless otherwise stated, the meaning of these terms is defined as follows. Please be aware that the definitions given here are not intended to be universal; rather, the definitions are specifically tailored to fit the environment at MSI.

General Terms

  • Active Group - Group that has renewed or opened during the current calendar year.

  • Annual Renewal - The process by which research groups renew their access to MSI’s HPC services for the next calendar year. Each Principal Investigator or Group Administrator is responsible for renewing their group’s access, and annual renewal is required to maintain access. Notices are generally sent out on November 1.

  • Class Account - A temporary group, generally 1 semester, that is provisioned for instructors that would like to make use of compute resources.

  • Group Admin - A member of a Principal Investigator’s group who has privileges to carry out administrative functions for the group, such as adding new users and giving group members access to HPC resources. The PI may assign this role to any MSI user in their group.

  • Group Member - User whose account is associated with an active group.

  • MSI Project - Group owned by a PI with access to MSI systems.

  • MyMSI - Web portal for managing MSI Projects, memberships, storage allocation, annual renewals.

  • PI - Principal Investigator, the primary individual responsible for access and research conducted using MSI systems, generally a faculty member.

  • Open OnDemand (OOD) - Web resource for accessing compute resources

  • Regulated Project - A group that is verified to have completed a data use agreement (DUC) to work with protected datasets.

  • Restricted Project - Generally a group that has access to compute resources on our HIPAA compliant cluster Blackwell.

  • Standard Project - Default project type

  • Tutorials - Series of instructor-led training material that is presented every semester.

Compute Terms

  • Compute Node - A compute resource typically CPU-only.

  • CPU, Task, Thread, Worker - Single unit of compute, e.g., 1 Node on Agate has 128 tasks available from 2 physical CPUs that each have 64 threads.

  • GPU Node - A compute resource that also has one or more GPUs.

  • Login Node - ‘Front Door’ node, intended for interfacing with the Job Scheduler and navigating filesystem.

  • Memory (RAM) - Volatile short-term memory; a single node is limited by the amount of memory physically installed.

  • Node - Single ‘computer’ that is made up of CPUs, Memory, GPUs; A single job can have one to many nodes associated.

  • Temp Disk, Local Scratch - Storage that is local to the physical node. - /tmp - /scratch.local/

Storage Terms

  • Allocation - The resources that are provided to each PI group free of charge each year. Groups will receive an allocation of storage (GBs/TBs) each year, and must submit their Annual Renewal form each year to retain access to allocated resources. With additional justification, additional resources can be purchased for the PI group, called Purchased Storage.

  • Bucket - A ‘directory’ in Tier 2, as an object-based filesystem, it does not have the same concept of directories as on Tier1.

  • File Quota - Maximum individual file limit.

  • Global Scratch - Temporary scratch space that is available to all users for large bursts of data. - /scratch.global/

  • Globus - Browser-based interface for moving data between Tier 1 and Tier 2.

  • Home Directory - Private user directory. - /users/[0-9]/<UMN InternetID>

  • Local Scratch - Storage space available local to the compute resource. - --tmp <Amount>GB

  • Policy - Tier 2 permissions on a Bucket, by default only the owner has access.

  • Purchased Storage - Resources that may be purchased beyond the allocations given to each PI group. PIs may request to purchase dedicated resources, such as additional storage and hardware, to suit the needs of their research that normal allocations may not be able to meet. Additional justification may be needed for some purchases.

  • Project Space - File path to where the MSI Project can store data. - /projects/standard/<Project Name>/

  • Shared Space - Directory named ‘shared’ in the Project space. - /projects/standard/<Project Name>/shared

  • Snapshot - Storage backup service that runs on Tier 1, Standard Project spaces, and User directories. A snapshot is a read-only copy of data stored at a particular point in time, maintaining the same permissions for each file at that point in time. You can also snapshot while running Virtual Machines (VM). - cd .snapshot

  • Storage Quota - Maximum storage limit.

  • Supplemental Storage Request - Storage increase request filled out on MyMSI to increase storage limit for the calendar year.

  • Tier 1 - Main storage resource available to groups. The file system is available (exported) on all of MSI’s computational resources, so that users will see the same files and directories no matter what system they are using.

  • Tier 2 - An object storage system in place for storing inactive data, supporting data lifecycles and workflows, and facilitating data management and sharing. Second tier storage is ideal for those looking to share data with colleagues around the world, including some storage through Amazons S3 (Simple Storage Service) interface.

Slurm Terms

  • Batch Job - Shell script that is placed into the queue and will run once it becomes eligible, not in real-time.

  • Features, Constraints - Feature tags applied to differentiate between generations of compute node types.

  • Interactive Job - Compute resources that are used in real-time.

  • Partition - The scheduling partitions used to efficiently allocate MSI hardware and resources to different types of jobs. The job partitions on our systems manage different sets of hardware, and have different limits for quantities such as walltime, available processors, and available memory. A partition is aa collection of Nodes that can be requested following the resource definitions set by the Scheduler.

  • Job Array - Type of Slurm job where the same task is executed in many jobs with slight changes based on the index specified.

  • Job Script - Shell script that requests and executes commands on compute resources.

  • Queue - Schedule managed by the Scheduler utility, all jobs ‘sit’ in the queue as they become eligible to run based on job size and prioritization.

  • Scheduler - Referencing the ‘Slurm Scheduler’, the simple Linux utility for resource management.

  • Walltime - Timelimit specified for a job; how long should the job have access to these compute resources?

Software Terms

  • Module - A package that is preinstalled on the cluster and can be made available to the userspace by loading it with the command module load <module name> such as module load matlab. Users may also create private software modules of software installed within their home or group directories.

  • Package - individual piece of software; not necessarily a module.