Tru Huynh <[email protected]> writes: > Hello, > > The nodes are booting diskless(warewulf-3.5-1), and use NFSv3 for everything, > beegfs(2015.03.r7-el6) for scratch. > I am running slurm-14.11.10 and I have a memory reporting issue: > > [tru@visu1 ~]$ sinfo -Nel > Tue Dec 15 14:48:48 2015 > NODELIST NODES PARTITION STATE CPUS S:C:T MEMORY TMP_DISK > WEIGHT FEATURES REASON > c[001-040] 40 main* idle 16 2:8:1 1 0 > 1 (null) none > c[041-049,054-056] 12 main* idle 20 2:10:1 1 0 > 1 (null) none > c050 1 main* drained 20 2:10:1 1 0 > 1 (null) nvidia > c[051-053] 3 main* mixed 20 2:10:1 1 0 > 1 (null) none > > The MEMORY reports the default amount of memory, instead of the actual memory > size: > Log on the nodes: > [2015-12-15T12:45:00.306] No specialized cores configured by default on this > node > [2015-12-15T12:45:00.306] Resource spec: Reserved system memory limit not > configured for this node > [2015-12-15T12:45:00.310] slurmd version 14.11.7 started > [2015-12-15T12:45:00.314] slurmd started on Tue, 15 Dec 2015 12:45:00 +0100 > [2015-12-15T12:45:00.315] CPUs=16 Boards=1 Sockets=2 Cores=8 Threads=1 > Memory=129167 TmpDisk=64583 Uptime=119 CPUSpecList=(null) > > [tru@c001 ~]$ df -HP > Filesystem Size Used Avail Use% Mounted on > none 68G 1.2G 67G 2% / > tmpfs 68G 0 68G 0% /dev/shm > 10.11.3.10:/home 1.1T 865G 235G 79% /baycells/home > 10.11.3.10:/scratch 5.5T 867G 4.7T 16% /baycells/scratch > 10.11.3.2:/c6/shared 322G 231G 92G 72% /c6/shared > 10.11.3.2:/c6/shared/rpm 30G 15G 15G 51% /c6/shared/rpm > beegfs_nodev 6.2T 246G 5.9T 4% /mnt/beegfs > > slurm.conf: > ... > SelectType=select/cons_res > SelectTypeParameters=CR_CPU > ... > NodeName=c0[01-40] Sockets=2 CoresPerSocket=8 ThreadsPerCore=1 State=UNKNOWN > NodeName=c0[41-56] Sockets=2 CoresPerSocket=10 ThreadsPerCore=1 State=UNKNOWN > PartitionName=main Nodes=c0[01-56] Default=YES MaxTime=INFINITE State=UP > ... > > Do I have to put RealMemory in the config file?
Yes. > Thanks > > Tru -- Dr. Loris Bennett (Mr.) ZEDAT, Freie Universität Berlin Email [email protected]
