| Version 4 (modified by , 12 years ago) ( diff ) |
|---|
Bluegene cluster
The CRC's wikipage on the bg/p system is found here. As of 06/2010, the cluster has 1,024 nodes, with each node having a quad-core PPC450 with 8 MB of cache and 2 GB of RAM. So, using all cores on a node results in ½ GB of RAM per core.
As discussed on the RunningOnBluegene page, jobs must be submitted in sets of 64 nodes (256 cores). Hence, a maximum of 16 jobs running concurrently is possible at any one time.
FAQ
What to do with core files
When a process finishes with an error on bg/p, if compiled with -g, each processor will dump a core. file. These files contain information that can be used to track down where the error occurred with addr2line. E.g.:
$> addr2line -e myprog < core.NNNN | tail
where myprog is the executable invoked with mpirun and core.NNNN the specific core dump file you wish to query. At the end of the core file will be some HEX addresses which, when the executable is supplied, allows addr2line to convert into line numbers from the source code.
