Shooting straight with Oakley stats

Communications Specialist
,
OH-TECH
Wednesday, February 13, 2019 - 10:30am

OSC’s Oakley Cluster, decommissioned in December 2018, had an impressive run at the Center, with over 10 million jobs completed since January 14, 2012. The $4.1 million HP-built, Intel® Xeon® processor-based supercomputer doubled OSC’s computing power at the time, using half as many nodes as the Glenn Cluster.

 

Named for legendary sharp shooter and Ohio pioneer Annie Oakley, OSC’s Oakley cluster had an impressive run in its six years at OSC. The Oakley Cluster was the first OSC cluster named for a female, followed shortly by Ruby in 2015. Oakley was a particular boon to those using graphics processing units (GPUs) as it tripled the number previously at the center.

 

Oakley’s decommissioning marks a major milestone for OSC. Staff pulled together some statistics on Oakley across its entire life:

 

Jobs Summary:

  • 380,875,520 total core hours
  • 10,822,515 total jobs run (11,428,989 job ids assigned)
  • 1,202 distinct projects ran jobs
  • 76.3% average utilization

 

Timing Tidbits:

  1. Physical delivery: 12/28/2011
  2. First day jobs could be run:  1/14/2012
  3. Last day jobs could be run: 12/18/2018
  4. Total days available for jobs: 2,531

 

Specific Job/Core Stats:

  • 6,212,175 jobs (57.4% of total jobs on Oakley) used only 1 processor
  • 9,890,036 jobs (91.3% of total jobs on Oakley) used only 1 node (12 or less processors)
  • 271 jobs used more than the general parallel queue 1,500 core limit
  • 72 jobs used more than the default per user 2,040 core job limit
  • Two  jobs used 8,292 cores

 

Specific Job/Time Stats:

  • 73.8% of jobs ran in less than an hour
  • Longest running job was 839 hours
  • Average queue time across all jobs: 3 hours 0.6 mins
  • 55.7% of all jobs waited less than 10 mins in the queue

 

Specialized Node Stats:

  • 194,039 jobs run with GPU requested feature (1.8% of total)
  • 2,069 jobs ran on HugeMem node
  • 3,801 jobs ran requesting the BigMem nodes

 

Specific Software Stats:

  • Amber software used in the most # of jobs (1,040,699 / 9.6% of total)
  • VASP software used the most core hours (33,795,502 / 8.8% of total)
  • 328 distinct system-level software programs run at least once

 

Specific Interesting Jobs / Accounts:

  • Most active account by both core-hours and job count belonged to Le-Sheng Bai from Byrd Polar, with 16,394,986 core hours (4.3% of total core hours on Oakley) and 1,251,900 jobs (11.5% of total jobs on Oakley).
  • First job by a non-OSC staff member was run by Daniel Paquet, a grad student from Dr. Sadayappan’s lab at the University of Cincinnati.
  • Last job run was by Supriyo Chakraborty under Dr. Shawn Midlam-Mohler from The Ohio State University.