I recall there was a built in tool named seff (slurm efficiency), not sure if it is still maintained, but here is a link from my bookmark collection related to some other slurm efficiency tool: rse. princeton. edu/2020/01/monitoring-slurm-efficiency-with-reportseff/ On the automation part, it would be pretty easy to do regularly(daily?) stats of jobs for that period of time and dump them into an sql database. Then a select statement where cpu_efficiency is less than desired value and get the list of not so nice users on which you can apply whatever warnings/limits you want to do.
Cheers, Cristian On Wed, Oct 11, 2023 at 10:59 AM Loris Bennett <loris.benn...@fu-berlin.de> wrote: > Diego Zuccato <diego.zucc...@unibo.it> writes: > > > IIUC Loris is looking for a method to do that automatically. > > Since there is a "used memory" metric, a corresponding "wasted memory" > > (or "wasted CPU") one with corresponding weights in priority > > calculations could be useful. > > Yes, an automated solution is indeed what I am after. > > A "naughty" list wouldn't really help, since there is quite a spectrum. > I would also need at least a "naive" list, an "erratic" list, and > probably a thankfully fairly short "recalcitrant" list. > > Cheers, > > Loris > > > Diego > > > > Il 11/10/2023 09:54, Williams, Gareth (IM&T, Black Mountain) ha scritto: > >> Hi Loris, You could add them to a naughty list (please use a better > >> term...) and apply a fixed commensurate priority penalty. That is > >> simple (and can apply to a range of unpreferred behaviour) and > >> provides a clear motivation to change. Could be done with QOS unless > >> you already use that in a conflicting way. > >> Gareth > >> Get Outlook for Android <https://aka.ms/ghei36> > >> ------------------------------------------------------------------------ > >> *From:* slurm-users <slurm-users-boun...@lists.schedmd.com> on > >> behalf of Loris Bennett <loris.benn...@fu-berlin.de> > >> *Sent:* Wednesday, October 11, 2023 5:26:38 PM > >> *To:* Slurm Users Mailing List <slurm-users@lists.schedmd.com> > >> *Subject:* [slurm-users] Fairshare: Penalising unused memory rather > >> than used memory? > >> Hi, > >> We have an increasing number of users who are unable (or unwilling) > >> to > >> estimate their memory requirements accurately. We include memory in our > >> calculation of fairshare usage and thus penalise everyone who requests > >> large amounts of memory, whether it is needed or not. > >> Therefore I would be interested in knowing whether one can take into > >> account the *requested but unused memory* when calculating usage. Is > >> this possible? > >> Cheers, > >> Loris > >> -- Dr. Loris Bennett (Herr/Mr) > >> ZEDAT, Freie Universität Berlin > >> > -- > Dr. Loris Bennett (Herr/Mr) > ZEDAT, Freie Universität Berlin > >
