link to homepage

Institute for Advanced Simulation (IAS)

Navigation and service

New GPFS scratch filesystem

A new GPFS scratch file system, based on IBM GPFS Storage
Server (GSS) infrastructure will be taken into production to
-) double the performance,
-) achieve extreme data integrity,
-) accelerate rebuild times with minimal performance impacts,
-) and thereby enhance data protection.

Because of the huge amount of data and the enormous number of
files in $WORK it is not reasonable to pre-copy all data into
the new file system. Your active help is needed to copy the data
you want to keep, using the standard Unix command "cp".

Please read the following work plan carefully and follow it

1) Since August 2013 the new optimized file system referenced by

the environment variable $WORKNEW is available.

It is recommended to start using the $WORKNEW file system
for scratch data at the HPC systems (JUQUEEN, DEEP).
Quotas for projects/groups are preserved during the transition
from $WORK to $WORKNEW.

2) On September 16, 2013, the $WORKNEW file system will become
the standard scratch file system and the environment variable
$WORK will be switched to reference this new filesystem.
So $WORKNEW and $WORK reference the same (new) file system.
The old file system will now be only accessible via
the environment variable $WORKOLD and it is set to READ-ONLY!

Jobs referencing data which were not yet been transferred, will
have to be changed.

August 2013 16. Sept. 2013
/worknewrw$WORKNEW-->/workrw$WORK, $WORKNEW

3) On January 2014, the file system referenced as $WORKOLD will
be deleted as it should be empty after 90 days by the normal
cleanup routine.

Some further remarks:

- At JUROPA the environment variables $GPFSWORK, $GPFSWORKNEW,
and $GPFSWORKOLD will be set on the GPFS nodes accordingly,
so you might need to adapt your scripts.

- As before, data that are older than 90 days and not accessed
for 90 days will be deleted automatically as well as empty
directories are already deleted after 3 days automatically.

- As before there is no backup for data stored in

Thank you in advance for your help. We are firmly convinced
that the planned rearrangements will make access to data on
the GPFS scratch file system more effective.

If you need some extra information please feel free to contact

SC-support at Juelich Supercomputing Centre
Phone: +49-2461-61-2828