1 / 10

Nikhef Multicore Observations

Nikhef Multicore Observations. Jeff Templon Multicore TF #2 2014.01.28. Data set. Jobs running on 26 and 27 january. Waiting time . Short pilots. Submission to short queue (4 hrs ) helped multicore throughput a lot. Production pilots.

joella
Download Presentation

Nikhef Multicore Observations

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Nikhef Multicore Observations Jeff Templon Multicore TF #2 2014.01.28

  2. Data set • Jobs running on 26 and 27 january Jeff Templon, Multicore TF #2

  3. Waiting time Jeff Templon, Multicore TF #2

  4. Short pilots • Submission to short queue (4 hrs) helped multicore throughput a lot Jeff Templon, Multicore TF #2

  5. Production pilots • Still could win some ground here with walltime limits ATLAS prod submits normally to ‘atlas’ or ‘medium’ queues: 52 and 38 hours resp. Jeff Templon, Multicore TF #2

  6. Multicore memory Jeff Templon, Multicore TF #2

  7. Memory and maui • Maui checks mem; recall we use pvmem • Maui assumes each proc needs ‘pvmem’ • Submit filter assumes “multicore” : 8 cores = 8 times mem … • Hence ATLAS was “asking for” 256 GB per job!!! • Changed by hand to 32 GB per job; seems ok (jobs consume 16) Jeff Templon, Multicore TF #2

  8. Maui & Memory II • My preferred solution: • Ask for nr cores: assumes independent cores • Hence pvmem = 4096mb • Need true multicore? Ask for mem too • But then Maui will over-ask because it does not understand multicore • “Not a problem” at moment? All ATLAS jobs use the indep. Core model Jeff Templon, Multicore TF #2

  9. Holding onto cores Jeff Templon, Multicore TF #2

  10. Holding onto cores • ATLAS doesn’t • Normal fair share fluctuations -> lose cores • Probably need some dynamic partitioning to “multicore” segment. Dynamic means others can use it for single-core work when ATLAS workload disappears. Jeff Templon, Multicore TF #2

More Related