Hi,
right now we are using torque+maui and do not have any shared home directories between workernodes and frontends. However in preparation for testing SLURM we'd like to move to a shared /home directory, in doing so we'd like to have the /home as a smaller volume (10TB 3-way replication) and then some say /scratch space in each WN that can be used for the jobs during execution for temporary space. Right now with no shared home PBS stages in the files from CREAM CE to /home/user/... and performs the computation in there staging the output back to CREAM and/or storage. Now we'd like it to not do the computation in /home, but instead move elsewhere.
We'd also like this to be CREAM reconfigure safe etc so probably it's something to be configured on the workernodes, but what does the wrapper script use to determine where it runs? Can sites who do use shared home directories describe how they've solved it? I'd really like to have for local users shared homes and for grid usage to trash the scratch area on each node and if at all possible to avoid stageins/outs as that's not supported by slurm.
Mario Kadastik, PhD
Researcher
---
"Physics is like sex, sure it may have practical reasons, but that's not why we do it"
-- Richard P. Feynman
