|Target usage: CELL cluster|
|System information: see prickly system page in web portal|
|System status: see prickly status page|
|Real time system data: see Ganglia monitoring page|
|Full list of SHARCNET systems|
Pickly was decommissioned as of June 2012.
This is a heterogeneous cluster comprised of 4 x86_64 multicore nodes and 8 Cell multicore nodes, connected via gigabit ethernet. The x86_64 nodes are dual-socket quad-core Xeon CPUs @ 2.50 GHz with 8GB RAM. The Cell nodes are dual-socket PowerXCell 8i CPUs @ 3.2 GHz with 16GB RAM.
NOTE: there is no automated job scheduler on prickly - users must email email@example.com to request a reservation for nodes.
For example (you'd want to be assigned the particular node to login to by contacting firstname.lastname@example.org), to login to pri05 remotely:
[user1@localhost]$ ssh prickly.sharcnet.ca [user1@prickly ~]$ ssh pri05 [user1@pri05 ~]$
At this point you should check to see who else is on the system, and whether or not anything is running, ie.
merz@pri05 ~]$ who merz pts/0 2009-03-26 09:59 (prickly.prickly.sharcnet) ^^^ only I'm logged in [merz@pri05 ~]$ uptime 10:00:18 up 62 days, 17:37, 1 user, load average: 0.00, 0.00, 0.00 ^^^ load = 0.00 means no jobs are running
You can also use the 'top' command to see what is running.
The Cell SDK can be found on the Cell nodes at:
Users can copy it into their work directory and then proceed with compiling and running the examples and demos.
The IBM XL compilers should be on your path if you're logged into one of the Cell nodes. It is recommended that users use this compiler, eg.
ppuxlC ppuxlc++ ppuxlc_r ppuxlf ppuxlf2003_r ppuxlf90_r ppuxlf95_r ppuxlc ppuxlC_r ppuxlc++_r ppuxlf2003 ppuxlf90 ppuxlf95 ppuxlf_r spuxlC spuxlc spuxlc++ spuxlf spuxlf2003 spuxlf90 spuxlf95
One can find the compiler documentation on the Cell nodes at:
Fortran: /opt/ibmcmp/xlf/cbe/11.1/README C/C++ : /opt/ibmcmp/xlc/cbe/10.1/doc/en_US/pdf/
Open MPI 1.3.1 for QS22
Users may wish to use an MPI compiled with the XL compilers, if so they should add the following to their shell environment (ie. the end of their ~/.bashrc file):
#host specific configs case `echo $CLUSTER` in prickly) export PATH="/work/merz/lib64/openmpi-1.3.1-xl/bin/:$PATH" ;; *) ;; esac
The above can be extended to configure your environment for other specialized SHARCNET clusters.
At this point, users will want to consult the Open MPI FAQ for further details about how to run MPI jobs.