Hints for using the IBM-Regatta-Power6 system at KISTI
Since 2009, KISTI (Korean Institute of Science and Technology) is running an IBM-Regatta system with Power6 CPUs and 24 batch nodes with up to 64 cores. It can be accessed via the interactive node gaiad.ksc.re.kr (150.183.146.24). The PALM scripts have been adjusted to be used on that system in March 2011 (r693). Batch jobs can be submitted directly from the interactive node or from any remote computer.
Submission of batch jobs from remote (i.e. from a local host) requires the following actions:
- Since KISTI's firewall does not allow outgoing ssh/scp connections, you have to ask KISTI about opening the outgoing ssh/scp port 22 for the IP-address of your local computer.
If the KISTI firewall is not opened properly, PALM jobs started from remote cannot transfer back results and job protocols. Also, restart runs from remote do not work.
- KISTI will open the firewall only for the interactive node gaiad. Since PALM jobs are running on the batch nodes, they have to redirect all ssh/scp transfer via gaiad. The PALM scripts do this automatically using internal ssh from the respective batch nodes to the interactive node, which requires you to create an ssh-key. Login on gaiad and execute:
ssh-keygen -t dsa cat id_dsa.pub >> authorized_keys
In order to make it work (i.e. to generate the host keys), you have to login once from gaiad to one of the batch nodes, e.g. gaia13 and vice versa:ssh <replace by user username on gaia>@gaia13 # login from gaiad to gaia13 ssh <replace by user username on gaia>@gaiad # login from gaia13 to gaiad exit exit
If you omit this step, jobs will abort with a message like "host key identification failed".
- mbuild, mrun, and subjob know this machine under the name ibmkisti, so you should have (at least) a block "ibmkisti parallel" in your configuration file (.mrun.config). You can find a default version for this block in the svn-repository under /palm/trunk/SCRIPTS/.config_block_for_ibmkisti.
Users from the Department of Atmospheric Sciences at Yonsei University, who like to run remote jobs on KISTI's IBM from local machine "climate0" at Yonsei Univ., can find a complete configuration file under /palm/trunk/SCRIPTS/.mrun.config.yonsei2011.