OSC Reflects on Recently Decommissioned Ruby Cluster

Communications Specialist
,
OH-TECH
Thursday, November 12, 2020 - 2:45pm (updated Friday, November 13, 2020 - 10:17am)
Ruby Cluster

In a year already full of twists and turns, the Ohio Supercomputer Center’s (OSC) decommissioning of their Ruby Cluster last month served as another notable event of 2020. The $1.5 million system was built in 2014 and ran more than 2 million jobs over six years.

OSC typically chooses to name their systems after a notable individual with ties to Ohio, and the Ruby Cluster was no exception. The cluster was named in remembrance of the late Ruby Dee, a Cleveland-born actress, author and civil rights advocate. Dee actively participated in marches and speeches for racial equality and was awarded the American National Medal of the Arts from the National Endowment of the Arts in Washington, D.C.

During its time in use, the Ruby Cluster aided researchers across a variety of disciplines. The system supported research focused on mechanical and aerospace engineering, protein structure predictions, computational fluid dynamics, and numerous other topics important in academia and industry alike.

For a better perspective of the Ruby Cluster’s features, take a look at the statistics below put together by OSC staff:

Jobs Summary:

  • 157,226,305 total core hours
  • 116,116 total GPU hours
  • 2,018,265 total jobs run
  • 147 distinct projects ran jobs

Timing Tidbits:

  • Physical delivery: 9/24/2014
  • First day jobs could be run: 3/16/2015
  • Last day jobs could be run: 10/29/2020
  • Total days available for jobs: 2,530

Specific Job/Core Stats:

  • 326,096 jobs (16% of total jobs on Ruby) used only 1 processor
  • 1,258,161 jobs (62% of total jobs on Ruby) used only 1 node (20 or fewer processors)
  • 31 jobs used more than the general parallel queue 800 core limit
  • 2 jobs used 4,800 cores

Specific Job/Time Stats:

  • 86% of jobs took less than an hour to run
  • Longest running job was 192 hours
  • Average queue time across all jobs: 2 hours and 40 minutes
  • 56% of all jobs waited less than 10 minutes in the queue

Specialized Node Stats:

  • 22,742 jobs run with GPU nodes (1% of total)
  • 2,504 jobs ran on HugeMem node

Specific Software Stats:

  • VASP software used the most core hours (45,481,968 / 29% of total)
  • VASP software ran the greatest number of jobs (587,275 / 29% of total)
  • NAMD software used the most GPU hours (26,313 / 23% of total)
  • 195 distinct system-level software programs ran at least once

Specific Interesting Jobs/Accounts:

  • Most active account (by core-hours): Minkyu Kim, in Aravind Asthagiri’s group, with 13,065,539 core hours (8% of total core hours on Ruby) 
  • Most active account (by jobs): Szu-Chia Chien, in Wolfgang Windl’s group, with 330,395 jobs (16% of total jobs on Ruby)
  • First job by a non-OSC staff member: Ray Leto, run on 04/16/2014 on 2 nodes  
  • Last job: Trey Dockendorf, run on 10/28/2020 on 2 nodes