>>>>> In >>>>> <d968e1a61e2a924780f82fccc09474a6fe89323...@nspexmbx-c.the-lab.llnl.gov> >>>>> "Lipari, Don" <lipa...@llnl.gov> wrote:
> > -----Original Message----- > > From: owner-slurm-...@lists.llnl.gov [mailto:owner-slurm- > > d...@lists.llnl.gov] On Behalf Of Andreas Davour > > > Do I also have to set priorities and weights to accounts and users > > > using sacctmgr? > The only thing you assign to accounts and/or users using the > sacctmgr is the share value (aka "fairshare"). > > > Just to make it clear. Do I need to add entries in the db for > > > everything I want to factor in into the fairshare? Or do the schema > > > contain everything you can collect accounting data on? I'm not that > > > familiar with mysql so I'm not sure how to check. > All of the PriorityWeight* settings in the slurm.conf contribute to > the job priority calculation when using the priority/multifactor > plugin. The only relevant components in the database are > "fairshare" and QOS priority. > > > If I have understood correctly, I not only have to set the > > > PriorityWeightFaishare directive in slurm.conf, but also add weight to > > > accounts, clusters and fairshare elements in the database, right? > You cannot enter weights into sacctmgr. The only PriorityWeight* > settings in the slurm.conf file that act on anything from the > database is PriorityWeightFairshare and PriorityWeightQOS. > Instead you assign shares to each account/user that represents how > much of the machine they are entitled to use. > > I realized that some examples might be a good idea, illustrating my > > problems. I think I have enetered stuff in the db using sacctmgr but > > it still looks wonky. > [...] > > I expected atlm-000, job 52440, to have much higher FAIRSHARE number > > than job 53973?? > Try running to sshare -al command to get a fuller picture of how the > fairshare factor of each user/account was derived. By the way, none > of the usage numbers (raw, normalized, or effective) listed are > stored in the database. Instead they are stored locally to the > machine in the configured " StateSaveLocation". Ok. I think that was more clean that before. But, this is where I get confused. atlas 1 0.066667 148071114 0.618961 0.618961 0.001604 atlas atlm-000 5000 0.000515 9047 0.000038 0.004821 0.001525 klasm 10 0.666667 87219380 0.364591 0.364591 0.684496 klasm swegrid 5 0.666667 87219380 0.364591 0.364591 0.684496 Then they both run jobs for a while and it looks like this: atlas 1 0.066667 148164849 0.619054 0.619054 0.001602 atlas atlm-000 5000 0.000515 9047 0.000038 0.004822 0.001523 klasm 10 0.666667 87285127 0.364339 0.364339 0.684675 klasm swegrid 5 0.666667 87285127 0.364339 0.364339 0.684675 The FairShare column decrease as atlm-000 runs, meaning he will get lesser chance to run, since he is using up his shares, right? So, why is the FairShare column increasing for the user swegrid in the klasm account as he is chewing up his shares? /andreas -- Systems Engineer PDC Center for High Performance Computing CSC School of Computer Science and Communication KTH Royal Institute of Technology SE-100 44 Stockholm, Sweden Phone: 087906658 "A satellite, an earring, and a dust bunny are what made America great!"
