1 of 46

Introduction to Interactive HPC

Michael Milligan

Fall 2025

1

MN Supercomputing Institute

for Advanced Computational Research

© 2009 Regents of the University of Minnesota. All rights reserved.

2 of 46

What is Interactive Computing?

  • Software GUIs
  • Prototyping workflows
    • Design your workflow for a single node (multi-core) or small set of nodes
    • Discover and test new tools/concepts
    • Profile, optimize and debug
  • Data Visualization

2

Interactive Computing is about immediacy. When the human requests an action, the computer immediately responds. When the computer finishes a task, the human immediately perceives the result.

MN Supercomputing Institute

for Advanced Computational Research

© 2009 Regents of the University of Minnesota. All rights reserved.

3 of 46

Why Interactive HPC?

HPC is part of a larger research workflow

Dr. so-and-so needs to explore a huge datasetNeeds: interactivity, analytics tools, large memory/storage

Dr. such-and-such is preparing to create a cutting-edge simulation visualizationNeeds: interactivity, remote visualization, compute, bandwidth

A grad student is prototyping a new algorithmNeeds: interactivity, dev and debug tools, time to iterate

MN Supercomputing Institute

for Advanced Computational Research

© 2009 Regents of the University of Minnesota. All rights reserved.

4 of 46

Tutorial Outline

  • Batch vs. Interactive Computing @ MSI
  • Getting Started with Interactive Resources
    • Connecting to MSI
    • Conscientious Computing
    • Prototyping Workflows
  • Hands-on
    • Remote desktops
    • Jupyter Notebooks

4

MN Supercomputing Institute

for Advanced Computational Research

© 2009 Regents of the University of Minnesota. All rights reserved.

5 of 46

MSI Access Overview

5

MN Supercomputing Institute

for Advanced Computational Research

© 2009 Regents of the University of Minnesota. All rights reserved.

6 of 46

Resource Allocations

ALL groups will have access to these basic resources:

  • High Performance Computing cluster access
  • High Performance Storage
    • 200 GB user home directory quota
    • Project storage accessible from most MSI systems
  • Access to interactive gateways
  • Access to technical support via helpdesk
    • access to expert consultants by referral

6

MN Supercomputing Institute

for Advanced Computational Research

© 2009 Regents of the University of Minnesota. All rights reserved.

7 of 46

Recent Changes

There have been some significant changes recently that may affect your usage:

  • Mesabi retirement
    • The mesabi cluster was decommissioned
    • Use agate for all your cluster computing needs
    • The mesabi login node is gone, connect to agate now
    • The mangi nodes were moved to agate, find them there
  • User home directories now separate from projects
    • Your $HOME is now in a separate /users/ space
    • Your $HOME quota is fixed and doesn’t affect project space
    • Find your project area (and old home directory/old files) in the usual location at /home/PROJECT/

7

MN Supercomputing Institute

for Advanced Computational Research

© 2009 Regents of the University of Minnesota. All rights reserved.

8 of 46

Survey of Options

8

MN Supercomputing Institute

for Advanced Computational Research

© 2009 Regents of the University of Minnesota. All rights reserved.

9 of 46

Batch Jobs

  • When should you use Batch Jobs?
    • Whenever possible! This is the traditional way to work in HPC
    • Share resources and be considerate of other researchers
  • What are the benefits of Batch Jobs?
    • Headless execution of automated processes
    • Long runtimes
    • Large core counts
    • A scheduler packs jobs in hardware to maximize utilization, reduce latency, etc.

9

MN Supercomputing Institute

for Advanced Computational Research

© 2009 Regents of the University of Minnesota. All rights reserved.

10 of 46

Interactive Options

  • OnDemand Portal
  • SSH terminal
  • Interactive batch (srun via the command line)
  • Galaxy portal (graphically compose workflows)
  • Specialized solutions
    • CITRIX for Windows apps

10

MN Supercomputing Institute

for Advanced Computational Research

© 2009 Regents of the University of Minnesota. All rights reserved.

11 of 46

OnDemand Portal

https://ondemand.msi.umn.edu

  • OnDemand Web Portal is:

One-stop access portal for interactive HPC services

    • Remote desktop access
    • Command line terminal
    • File browser
    • Launchers for common GUI scientific applications
    • JupyterLab server access
    • RStudio server access

11

MN Supercomputing Institute

for Advanced Computational Research

© 2009 Regents of the University of Minnesota. All rights reserved.

12 of 46

Interactive Command Line

ssh <username>@<resource>.msi.umn.edu

See: https://www.msi.umn.edu/content/connecting-hpc-resources

12

  • SSH is the traditional first point of access to MSI resources
  • Access agate login head node
  • Uses ssh program available on most personal computers

These “login nodes” are shared by many MSI users. They are NOT for heavy computational tasks. Appropriate uses include:

  • Browsing and editing files
  • Submitting jobs
  • Querying job status and other informational commands
  • Browsing software modules
  • Light testing or software compilation

MN Supercomputing Institute

for Advanced Computational Research

© 2009 Regents of the University of Minnesota. All rights reserved.

13 of 46

Interactive Batch with ‘srun’

  • Dedicated compute resources via the scheduler
  • Command line terminal running on the cluster
    • similar to a SSH session
  • Supports all usual Slurm options and commands
  • Best used with interactive partitions
    • have priority access for small interactive job requests
  • Job terminates when shell exits or connection broken

srun <resource requests> --pty bash

See: https://www.msi.umn.edu/content/interactive-queue-use-srun

13

Example:

srun -N 1 --ntasks-per-node=4 --mem-per-cpu=1gb -t 1:00:00 -p interactive --pty bash

MN Supercomputing Institute

for Advanced Computational Research

© 2009 Regents of the University of Minnesota. All rights reserved.

14 of 46

Web Portals: Galaxy

http://galaxy.msi.umn.edu

Galaxy is a web-based workflow manager, principally for bioinformatics/’omics research

14

  • Compose data analysis workflows in a unified data flow view
  • Workflows execute when requested on MSI compute clusters
  • Wide variety of bioinformatics and ‘omics applications available as tools

See https://www.msi.umn.edu/content/galaxy for additional details, especially regarding linking data into Galaxy

MN Supercomputing Institute

for Advanced Computational Research

© 2009 Regents of the University of Minnesota. All rights reserved.

15 of 46

Need Windows?

MSI no longer provides Windows application servers.

Select HPC applications that require Windows are available through partnership with OIT via CITRIX

See: �https://msi.umn.edu/about-msi-services/interactive-hpc/citrix for access to Apps.umn.edu

15

MN Supercomputing Institute

for Advanced Computational Research

© 2009 Regents of the University of Minnesota. All rights reserved.

16 of 46

Hands-On: Getting Connected

16

MN Supercomputing Institute

for Advanced Computational Research

© 2009 Regents of the University of Minnesota. All rights reserved.

17 of 46

Pre-requisites to connect

17

  • MSI services are only available to on-campus network addresses
  • Especially during the pandemic, many users are working off-campus
  • VPN clients and instructions are available from OIT for most operating systems and mobile devices
  • DUO 2-factor authentication is required for most MSI services
  • DUO is also required to connect to the campus VPN

MN Supercomputing Institute

for Advanced Computational Research

© 2009 Regents of the University of Minnesota. All rights reserved.

18 of 46

Sidebar: Windows�SSH and File Transfer

Current Windows 10+ computers will have ssh/scp available from the Windows terminal.

For users who desire a graphical ssh or file transfer client:

https://winscp.net/eng/downloads.php

Download and install both WinSCP and PuTTY*

*MSI recommends WinSCP and PuTTY for Windows users, but does not offer support for these programs. Contact your local IT department for assistance.

18

MN Supercomputing Institute

for Advanced Computational Research

© 2009 Regents of the University of Minnesota. All rights reserved.

19 of 46

Choosing a Cluster/Partition

See: https://msi.umn.edu/our-resources/slurm-scheduler/choosing-a-partition

  • Long running jobs and batch jobs run on Agate
  • MSI compute clusters are heterogeneous (e.g., up to 2TB Memory, GPUs, SSDs, etc.)
    • Think: match your needs to available configurations
    • Do: reserve the (CPU/RAM/GPUs/etc) that you need
    • Avoid: reserving much more (waste, long queue times) or �less than required (job doesn’t finish before walltime expires)

For interactive use, begin with the interactive partition and experiment from there.

19

MN Supercomputing Institute

for Advanced Computational Research

© 2009 Regents of the University of Minnesota. All rights reserved.

20 of 46

Interactive Services OnDemand

20

MN Supercomputing Institute

for Advanced Computational Research

© 2009 Regents of the University of Minnesota. All rights reserved.

21 of 46

OnDemand Walkthrough

21

MN Supercomputing Institute

for Advanced Computational Research

© 2009 Regents of the University of Minnesota. All rights reserved.

22 of 46

OnDemand Walkthrough

22

Services accessed through menu bar

MN Supercomputing Institute

for Advanced Computational Research

© 2009 Regents of the University of Minnesota. All rights reserved.

23 of 46

OnDemand Walkthrough

23

Return to landing page

Dashboard of running and completed interactive activities

Job list and composer app�(web interface panels to squeue and sbatch)

MN Supercomputing Institute

for Advanced Computational Research

© 2009 Regents of the University of Minnesota. All rights reserved.

24 of 46

OnDemand Walkthrough

24

File browser access

File browser:

  • browse/navigate/manage files
  • upload/download files up to few GB size
  • access different starting dirs if you have multiple groups/projects

MN Supercomputing Institute

for Advanced Computational Research

© 2009 Regents of the University of Minnesota. All rights reserved.

25 of 46

OnDemand Walkthrough

25

Command line terminal access

Command line terminal:

  • “SSH” in a web browser (almost)
  • Launch terminal on cluster headnode
  • Roughly equivalent to e.g.�ssh <username>@agate.msi.umn.edu
  • Also launch in specific directory via file browser pane

MN Supercomputing Institute

for Advanced Computational Research

© 2009 Regents of the University of Minnesota. All rights reserved.

26 of 46

OnDemand Walkthrough

26

Interactive Apps menu lists available application launchers

Each launcher leads to a session options page

Each session runs as an interactive job on a cluster resource

Launchers also available via the Interactive Sessions dashboard

MN Supercomputing Institute

for Advanced Computational Research

© 2009 Regents of the University of Minnesota. All rights reserved.

27 of 46

OnDemand Walkthrough

27

My Interactive Sessions

List job status for interactive sessions: queued, running, finished

Launch into a running session (usually opens new browser tab)

Access full menu of interactive launchers

MN Supercomputing Institute

for Advanced Computational Research

© 2009 Regents of the University of Minnesota. All rights reserved.

28 of 46

OnDemand Walkthrough

28

My Interactive Sessions

Delete a running or queued session to cancel the job and release compute resources early

Deleting a completed job removes it from this dashboard; no other effect

MN Supercomputing Institute

for Advanced Computational Research

© 2009 Regents of the University of Minnesota. All rights reserved.

29 of 46

OnDemand Walkthrough

29

Remote desktop sessions

VNC embedded in a web browser

Runs on compute node in selected cluster resource

These launchers start a remote desktop session and launch corresponding application

Jupyter and RStudio servers run on compute node

Web interface proxied through OnDemand

MN Supercomputing Institute

for Advanced Computational Research

© 2009 Regents of the University of Minnesota. All rights reserved.

30 of 46

Interactive Linux Desktops

30

Interactive desktop: noVNC session proxied through OnDemand server

Graphical fidelity will suffer on low-bandwidth connections but quality can be adjusted to maintain usability

Good connections allow graphically intensive apps, GPU accelerated graphics available on GPU nodes

Session persists until logged out on desktop - don’t forget to log out to release resources when done!

MN Supercomputing Institute

for Advanced Computational Research

© 2009 Regents of the University of Minnesota. All rights reserved.

31 of 46

OnDemand Walkthrough

31

Example launcher:

Jupyter server

Options to customize most aspects of the interactive job

  • cluster / partition
  • group (if multiple available)
  • resource requests
  • launcher-specific parameters (e.g. Python environment for Jupyter servers)

MN Supercomputing Institute

for Advanced Computational Research

© 2009 Regents of the University of Minnesota. All rights reserved.

32 of 46

OnDemand Walkthrough

32

Logout and server restart options

Session cleanup

OnDemand sessions run as scheduled jobs: �cancel jobs or use desktop “logout” to end jobs and release resources early, or allow them to close automatically when walltime expires

Logging out from this page clears browser login but otherwise has no effect on active OnDemand sessions

Rarely may need to use “restart server” option to recover from failures

MN Supercomputing Institute

for Advanced Computational Research

© 2009 Regents of the University of Minnesota. All rights reserved.

33 of 46

Session Clean-up and Resume

  • What is lost when the connection drops?
    • Session: if not expired, reconnect to the same session/node
    • Data: no loss
  • What if you End Job/Logout Desktop?
    • Session: Terminated; Reconnect and start fresh
    • Data: Scratch data is deleted periodically* (clean up after yourself to be sure)
    • No loss for data in group home
  • How do you reconnect?
    • Login again through OnDemand Interactive Sessions dashboard

33

MN Supercomputing Institute

for Advanced Computational Research

© 2009 Regents of the University of Minnesota. All rights reserved.

34 of 46

Interactive Jobs with �srun

34

MN Supercomputing Institute

for Advanced Computational Research

© 2009 Regents of the University of Minnesota. All rights reserved.

35 of 46

Selecting Resources

  • SSH to the cluster headnode:
    • ssh <username>@agate.msi.umn.edu
    • Launch interactive terminal in OnDemand
  • Queue an Interactive job on the cluster:
    • srun <job options> --pty bash

35

MN Supercomputing Institute

for Advanced Computational Research

© 2009 Regents of the University of Minnesota. All rights reserved.

36 of 46

Selecting Resources

Options to srun specify job requirements:

    • srun [options]
      • --time=8:00:00 job can run for up to 8 hours
      • -N 1 --cpus-per-task=4 job requests 4 CPUs on one node
      • --mem=2500m job requests 2500 MB (2.5GB) of RAM
      • -p interactive
      • --pty bash job must specify shell program to run
    • Enable graphics via X-tunneling (--x11)
      • not available in web terminals (OnDemand, Jupyter, etc)

36

MN Supercomputing Institute

for Advanced Computational Research

© 2009 Regents of the University of Minnesota. All rights reserved.

37 of 46

Selecting Resources

Refer to SLURM partitions reference page:�https://msi.umn.edu/our-resources/slurm-scheduler/slurm-partitions

Example - basic interactive session with 4 CPU cores, �8 GB RAM, 4 hours runtime:

srun -N 1 --ntasks-per-node=4 --mem-per-cpu=1gb -t 1:00:00 -p interactive --pty bash

Example - interactive session for machine learning with 1 A40 GPU, 12 CPU cores, 1 hour runtime:

srun -n 12 -t 1:00:00 -p interactive-gpu --gres=gpu:a40:1 --pty bash

37

MN Supercomputing Institute

for Advanced Computational Research

© 2009 Regents of the University of Minnesota. All rights reserved.

38 of 46

Navigating Home and Project Spaces

  • Start location will be $HOME (something like /users/7/YOUR_ID)
  • Most work will happen in project space

cd $MSIPROJECT�or simply type msiproject

  • If you are attached to multiple project groups you may need to select the correct one

newgrp PROJECTNAME�srun -A PROJECTNAME <more options>

Similar effect to selecting group from dropdown in OnDemand

38

MN Supercomputing Institute

for Advanced Computational Research

© 2009 Regents of the University of Minnesota. All rights reserved.

39 of 46

Software Modules

  • See all modules available
    • module avail
  • Load a module (adds commands to your shell)
    • module load python3
  • Run the software from the module
    • ipython

39

MN Supercomputing Institute

for Advanced Computational Research

© 2009 Regents of the University of Minnesota. All rights reserved.

40 of 46

Clean up your session

  • Delete the session (choose one)
    • Exit out of all connections
    • scancel <job ID>
  • What happens on disconnect/delete?
    • All computation stops
  • How to resume?
    • Start with fresh session
    • Load intermediate files (if you created them)

40

MN Supercomputing Institute

for Advanced Computational Research

© 2009 Regents of the University of Minnesota. All rights reserved.

41 of 46

Questions?

These slides and the hands-on guide will be posted to http://msi.umn.edu (see the “MSI Tutorials” section)�

41

MN Supercomputing Institute

for Advanced Computational Research

© 2009 Regents of the University of Minnesota. All rights reserved.

42 of 46

Extra Material

42

MN Supercomputing Institute

for Advanced Computational Research

© 2009 Regents of the University of Minnesota. All rights reserved.

43 of 46

Optional: SSH Keys

43

MN Supercomputing Institute

for Advanced Computational Research

© 2009 Regents of the University of Minnesota. All rights reserved.

44 of 46

Hands-On: Interactive Development and Data Interrogation with Jupyter

44

MN Supercomputing Institute

for Advanced Computational Research

© 2009 Regents of the University of Minnesota. All rights reserved.

45 of 46

JupyterLab Workspace

45

  • Rich tabbed multi-document workspace
  • File navigation able to access home and group directories
  • Reproducible, sharable computational stories in Jupyter Notebooks
  • View and edit text and other files directly in JupyterLab
  • Command line terminal available

MN Supercomputing Institute

for Advanced Computational Research

© 2009 Regents of the University of Minnesota. All rights reserved.

46 of 46

Additional Documentation

46

MN Supercomputing Institute

for Advanced Computational Research

© 2009 Regents of the University of Minnesota. All rights reserved.