The thread has somewhat been branched for the RAM requirements. Any useful 
comments on the #2-#4? I can probably summarize this by running over all 
compute nodes with scontrol show host, but that may not be too efficient...

On 25.04.2013, at 17:28, Mario Kadastik <[email protected]> wrote:

> 
> Hi,
> 
> I'm trying to get an overview of the state of the cluster. What I'd really 
> like to know is for example:
> 
> 1) compute nodes online 
> 2) compute cores online
> 3) compute cores allocated
> 4) distribution of job sizes currently running (and queued possibly)
> 5) list of nodes that are down/draining and reason
> 
> out of those #1 and #5 can be gotten from sinfo command with sinfo -Nle -p 
> main, which shows nodes and their states with reasons. 
> 
> However I cannot find right now quickly how to find out how many cores in 
> total are online (in theory it's nodes up * cpu count / node summed for each 
> node type) and even more crucial is how many cores are actually used and by 
> what size jobs. Today I was really tearing my hair out as 99% of the time we 
> use single core jobs and on my ca 4300 cores I only saw ca 1800 jobs with 
> 6000 in queue. As it came out a user had submitted 5 jobs with subtasks. Four 
> had 100 subtasks and one had 2000 nicely accounting for the missing jobs. 
> However I would really appreciate some summary view of the cluster. Is it 
> already available in sinfo, sstat, scontrol commands? If not, does anyone 
> have a good script that gathers the info together efficiently and lists it. 
> 
> It'd have to be text only as all nodes are headless and I'd prefer to get the 
> overview in a nice summary in shell. 
> 
> Thanks,
> 
> Mario Kadastik, PhD
> Researcher
> 
> ---
>  "Physics is like sex, sure it may have practical reasons, but that's not why 
> we do it" 
>     -- Richard P. Feynman

Mario Kadastik, PhD
Researcher

---
  "Physics is like sex, sure it may have practical reasons, but that's not why 
we do it" 
     -- Richard P. Feynman

Reply via email to