Logging on to the IBM Blue Gene/L System
The IBM Blue Gene/L uses a so called Front/End-Node or Login-Node for interactive access and submit of batch jobs.
Access
Access to the login node JUBL for users outside ZAM will only be possible by using the JUMP system as ssh-gateway.
A typical login from there looks like
-
ssh <userid>@jubl.zam.kfa-juelich.de
To submit and run jobs on the Blue Gene system the users must be known for queueing (mpirun-be) on the Service-Node but users are not allowed to login there. Any tries to do so will give cryptic results and must be cancelled by <Ctrl>-C.
Userid
The userid for the two High Performance Computers
JUBL and JUMP is the same because it's based on
the same LDAP registration entry. This means the login password will always be the same for both systems.
The user must have a valid JUMP userid when
requesting for a JUBL account. Not every
JUMP account automatically allows access to JUBL.
See also
"How to get computing resources on BG/L"
.
Home Data
JUBL and JUMP users have access to the same home data
in a common General Parallel File System (GPFS).
So data used while authentication is valid for each site especially the user's ssh keys and the stored public ssh keys for access from other systems.
For organization of the home data in respect to the two systems it is strongly recommended to add a new subdirectory in $HOME and to store the Blue Gene data there.
Attention: Please do NOT rename or reorganize the directory structure of the existing JUMP data because this will cause recall and backup activities for every changed file due to path or name modifications.
Login Shell
The Login Shell on each system points to system's favored shell:
-
JUMP -> ksh
JUBL -> bash
To avoid missmatches when logging on to the different systems it is recommended to do special settings within the .profile in respect to the system
-
if [ `uname` = "AIX" ]; then
... do JUMP stuff
elif [ `uname` = "Linux" ]; then
... do JUBL stuff
fi
last change 14.11.2007 |
Ulrike Schmidt | Print