[Ekhi-users] /scratch overloaded

[Inigo Aldazabal Mensa]

Hi all,

Remember to use /lscratch instead of /scratch when running one node
jobs. Most of you are using global /scratch and this is making it
*very* slow not only for working on it, but also it will make your
jobs run slower as they'll be waiting for file I/O instead of computing.

At this moment the cluster is fully loaded, so we can have 28*40=1120
jobs fighting for reading/writing to /scratch!! Bad thing as the
filesystem is not designed for this usage.

Please in your slurm scripts move the files back and forth /scratch
or $HOME to/from /lscratch, and run from /lscratch. Your jobs will
benefit also as you have SSD disks in /lscratch and not so in /scratch.

Any doubt about this please ask here so all can benefit.

Bests,

Iñigo



-- 
Iñigo Aldazabal Mensa, Ph.D.
Computing Service Manager / Scientific Computing Specialist
Centro de Física de Materiales (CSIC-UPV/EHU)
Paseo Manuel de Lardizabal, 5
20018 San Sebastian - Guipuzcoa
SPAIN

phone: +34-943-01-8780
e-mail: inigo.aldazabal en csic.es inigo.aldazabalm en ehu.eus
pgp key id: 0xDBCC8369