Dear Users,
due to the wide utilization of the sub-blocking technique on FERMI, we would like to inform you that is now implemented
such modality also for an allocated block of 128 compute nodes (i.e. bg_size=128).
You are now able to launch multiple runs in both 64 or 128 compute nodes, by dividing it in smaller blocks:
1) for bg_size=64, in 32, 16, 8, 4, 2, 1 compute nodes,
2) for bg_size=128, in 64, 32, 16, 8, 4, 2, 1 compute nodes,
Due to this new implementation, the subblock module has been changed. If you are already using the sub-blocking technique, some modifications have to be done on your job script, mainly regarding 1) the path of source file; 2) the export of the new environment variable N_BGSIZE.
NOTE: From tomorrow, your old subblock-job script will not work anymore.
More informations can be found at the web page
http://www.hpc.cineca.it/content/batch-scheduler-loadleveler-0#sub-block
Best Regards,
HPC-CINECA User Support