Difference between revisions of "Report from the SciComp Meeting on January 31"
From GlueXWiki
(Created page with "<pre> Follow-up: - HOW2019 in March - procmond - Process Monitor daemon - update and document - Discussion of farm graphs - job slots and hyperthreading, so 50% display is...") |
(wikitize) |
||
| Line 1: | Line 1: | ||
| − | + | from Sandy Phipott's notes: | |
| − | + | * HOW2019 in March | |
| − | + | * procmond - Process Monitor daemon - update and document | |
| − | + | * Discussion of farm graphs - job slots and hyperthreading, so 50% display is actually good! (*note* there would only be ~200 jobs running if every job consumed a whole node exclusively) | |
| − | + | * ifarm updates - 2 farm18 to become ifarm1801/02 | |
| − | + | * Slurm interactive and fast turn-around: | |
| − | + | ** priority queue ; to preempty a low priority queue | |
| − | + | ** production queue stays as-is, doesn't preempt others | |
| − | + | ** fairshare is still the basis | |
| − | + | * SWIF is updated to submit to Slurm | |
| − | + | ** testing is ready | |
| − | + | ** nodes start to move with February maintenance day (number to be determined) | |
| − | + | * 1.5x Lustre disk space is coming in March, with a Lustre upgrade to 2.10.6 | |
| − | + | ** installation has started, with a new metadata server | |
| − | + | * no more /work space; /work/clas clean-up (Friday - change projects to group-writable) | |
| − | + | ||
| − | + | ||
| − | + | ||
| − | + | ||
| − | + | ||
| − | + | ||
| − | + | ||
| − | + | ||
| − | + | ||
| − | + | ||
| − | + | ||
| − | + | ||
| − | + | ||
| − | + | ||
Latest revision as of 15:09, 5 February 2019
from Sandy Phipott's notes:
- HOW2019 in March
- procmond - Process Monitor daemon - update and document
- Discussion of farm graphs - job slots and hyperthreading, so 50% display is actually good! (*note* there would only be ~200 jobs running if every job consumed a whole node exclusively)
- ifarm updates - 2 farm18 to become ifarm1801/02
- Slurm interactive and fast turn-around:
- priority queue ; to preempty a low priority queue
- production queue stays as-is, doesn't preempt others
- fairshare is still the basis
- SWIF is updated to submit to Slurm
- testing is ready
- nodes start to move with February maintenance day (number to be determined)
- 1.5x Lustre disk space is coming in March, with a Lustre upgrade to 2.10.6
- installation has started, with a new metadata server
- no more /work space; /work/clas clean-up (Friday - change projects to group-writable)
