FAQ

Calculation

My job is `PENDING` with `Reason=Resources`

Check at your submit script, some mistypes can append.

When I submit a job, I got : `AssocGrpJobsLimit`

You reach your job quota on the requested partition.

When I submit an other job, I got : `AssocGrpCPUsLimit`

You reach your cpu quota on the requested partition.

Storage

My $HOME directory is full

Default quota is 100GB, do some cleanup, or ask to increase your quota if possible.

How many space I use in my $HOME ?

You can check this with :

quota -s

CPUs available on a node

How much cpus are allowed on a node ?

Use this command :

$ scontrol show node TESTNODE
NodeName=compute-0-4 Arch=x86_64 CoresPerSocket=12
   CPUAlloc=20 CPUErr=0 CPUTot=48 CPULoad=10.00 Features=rack-0,48CPUs
   Gres=ioperso:1
   NodeAddr=10.1.255.245 NodeHostName=compute-0-4 Version=14.11
   OS=Linux RealMemory=129180 AllocMem=115000 Sockets=4 Boards=1
   State=MIXED ThreadsPerCore=1 TmpDisk=260810 Weight=20528104
   BootTime=2015-09-02T11:30:42 SlurmdStartTime=2015-10-06T09:53:03
   CurrentWatts=0 LowestJoules=0 ConsumedJoules=0
   ExtSensorsJoules=n/s ExtSensorsWatts=0 ExtSensorsTemp=n/s

We can see 20 cpus allocated on a 48 cpus node.

Hyperthreading

I can’t launch an hyperthreaded job

By default, hyperthread is deactivated, to activate it :

$SBATCH --ntasks-per-core=2

Compilation

Which option to use to get a program usable on the entire cluster ?

Don’t use -x option like -xAVX, or use -axCORE-AVX2,AVX,SSE4.2

Which options match which CPUs

xeonv1 : -xAVX
xeonv2 : -xAVX
xeonv3 : -xCORE-AVX2
xeonv4 : -xCORE-AVX2
moonshot : -xSSE4.2

To get more informations about compiler flags : https://software.intel.com/en-us/articles/performance-tools-for-software-developers-intel-compiler-options-for-sse-generation-and-processor-specific-optimizations

MPI

I have to use `mpirun` or `srun`

To use mpi with slurm, you have to use srun

FAQ