ATTENDING

  • Jerry Sheehan, Josh Turner, Pol Llovet, Jonathan Hilmer, Thomas Heetderks

ABSENT

MINUTES: HPCAG Meeting #6 – (meeting presentation)

  1. Welcome - Jerry
  2. Hyalite Usage - Jerry & Pol
    • Overall
      • discussion: is Avg. CPU count reserved or actually used?
      • Pol: it is reserved
      • discussion: on my jobs, I reserve 8 CPUs even though I only need 2, because I need more RAM
      • Pol: there is a better way to manage reserved RAM in the scheduler [SLURM], we haven't yet implemented that additional complexity
    • Usage by College
      • Jerry: any thoughts about ways to use this data?
    • Usage by Department
      • discussion: CS & Math Depts have been competing to establish a Data Science Center-- when that gets going, we should have more users
    • Usage by Lab Group
    • Top Twenty Users
      • discussion: of course, most of the use from the top few users
      • Jerry: this is also how it is with the big players (ie most of Bluewaters useage by 20 or so)
    • Usage by Title
      • Jerry: we like that we have a good amount of student users
      • discussion: I think thats good!
      • discussion: no Math Science users?
      • Jerry: we're looking at other ways of promoting use: Data Science Competition, Jupyterhub, etc.
  3. GLOBUS: Usage last 12 months - Pol
    • Pol: you need a Globus-Plus account for transfer from personal to personal
    • Jerry: Globus allows anyone who uses Globus to be a transfering partner with our resources (in either direction)
    • Jerry: this is because of Globus handling identity
    • Jerry: they also have a nice tool for transfers (resumable, etc)
    • Jerry: and Globus is being used by many other campuses
    • Jerry: so how do we raise Globus awareness? --get more users?
    • discussion: send a letter (email?) to your cluster users
    • discussion: those who collect large amounts of data should need to transfer that data... can we target those users?
    • Jerry: we can focus on emerging science users or departments (cores)
    • discussion: I know that in Chemistry, they use a lot of big data
    • Pol: this use case somewhat overlaps with BOX, but for big transfers its so much better
    • Jerry: backup for labs with sensative/unbacked up data (BOX has a 15G file size limit)
    • Jerry: Globus has no bridge to BOX at this time (it does for Azure & S3)
    • Jerry: I have talked to BOX people to see if they are interested in Globus... we will see...
  4. Hyalite Software Usage - Pol
    • Pol: determining what users are running is hard-- an unsolved problem
    • Pol: you can just ask the user
    • Pol: we can see MATLAB license use and we've had a lot of success with the MATLAB license, but we don't have details about MATLAB CPU usage
      • Faculty/Staff/Lab usage: 148 licenses activated
      • Student usage: 574 licenses activated
    • discussion: good to see high student MATLAB usage
    • discussion: why do people load modules that they don't use?
    • Pol: people end up loading all the modules they think might need... then don't use many of them
  5. HPC Class Overview: CHMY591 - Pol
    • Pol: this is a small class (only 5), taught by Robert Szilagyi
    • Pol: there will be minimal impact on cluster
    • Pol: using scripts that Robert wrote to run class jobs, they don't need much shell use
    • Pol: they will use a special queue for their final jobs (limited nodes), with 8 CPUs per job
    • Pol: they are not really getting command line experience
    • discussion: so they're not really learning to use HPC
    • Jerry: we're just trying this approach, and wanting your feedback for this approach
    • Jerry: with more students, we told provost we would need additional HW for required classroom load
    • Pol: the other "issue" we had on this-- Robert likes AMD CPUs since his software prefers them, but we use Intel XEONs
    • Pol: if we add nodes for classroom use, we need to stick with our XEONs
  6. CHMY591 Jobs - Pol
    • Jobs will mostly be Gaussian
      • 8 CPUs, 16GB RAM
      • 72 hour job runtimes, max
    • Set up a custom class queue for runtime limit
      • class limited to 4 nodes
    • Impact on the total cluster usage will be minimal
  7. Hyalite Citations - Jonathan
    • Paper publisned by Ryan Snow (Physics) citing Hyalite:
      • "Calculations were performed on the Hyalite Research Cluster at Montana State University."
    • ACTION: we will get boilerplate posted for future Hyalite citations
  8. NSF CyberTraining Solicitation - Jerry
    • grant opportunity to provide unique training experiences
      • possibly training through Extended University
      • we are reaching out to vendors for possible partnership (Amazon AWS & Cisco)
      • obscurity of solicitation language makes it really though to know what they are looking for
      • an NSF seminar next week with grant PIs might tell us more?
      • we need to hear back from our vendors very soon with the quickly approaching Jan. deadline
      • reactions??
    • discussion: what about regular for credit courses in Data Visualization Analysis?
    • discussion: many students seem to struggle with Data Visualization

ACTIONS

  • Need to post boilerplate for Research Paper citations of Hyalite compute resources on Hyalite web pages

FUTURE AGENDA

  • Hyalite RAM Policy
  • Hyalite Storage Policy
  • Hyalite Expansion Update
  • CyberCANOE (display wall) installation