BCX: maintenance RESCHEDULED

[Carlo Cavazzoni]

Dear BCX users,


the maintenance of BCX has been rescheduled on

*** MONDAY the 22th of September ***


The maintenance operations will start at 8 a.m. and will end in late afternoon.

*** Four important change in the configuration will be done: ***

1) /scratch space will be increased of 24Tbyte
(DATA already present on /scratch will NOT be removed);

2) /scratch (GPFS) will be moved from GigabitEthernet to InfiniBand network,
this should increase the row performance by a factor of 10
(DATA already present on /scratch will NOT be removed);

3) new more strict memory limits will be set, so that applications trying
to allocate more than 90% of total virtual memory ( roughly 11Gbyte/node or 2.8Gbyte/core ) 
will be killed before compute node hangs;

4) default stack limit will be set to 10Mbyte to prevent shell applications (example tcl)
to use too much memory. If this limit is not enought for your application, you are free
to change it inside your job script before the application executable is run.
Example:
#!/bin/bash 
#BSUB -W 0:25
#BSUB -a openmpi
#BSUB -n 30
. /prod/modules/tcl_3.2/init/bash
module load openmpi/1.2.5/intel
module load acml/3.6.0--intel--10.1
ulimit -s unlimited
mpirun.lsf my_application.x
# -- end of example


best regards,
Carlo Cavazzoni

for any question, doubt or request of support
write to superc a cineca.it
-------------- parte successiva --------------
Un allegato HTML è stato rimosso...
URL: <http://list.cineca.it/pipermail/hpc-news/attachments/20080911/80d1764a/attachment.html>