OpenMX consumes memory and dies |
- Date: 2015/02/13 01:24
- Name: Artem Pulkin
<artem.pulkin@epfl.ch>
- Dear Taisuke,
The problem is somewhat similar to this one:
http://www.openmx-square.org/forum/patio.cgi?mode=view&no=1279
The input file contains ~70 atoms, band structure calculation. OpenMX runs successfully on older AMD cluster (1 node, 48 processes per node, 4 Gb per process) but fails on a new Intel cluster (1 node, 1 process per node, 64 Gb per process). It also fails if all 16 cores are used, 4Gb per core. The makefile adjustments are similar to the example bundled with OpenMX:
vsmp (Intel Xeon SMP cluster, Xeon E5-2690 0 @ 2.90GHz)
When I log into the node I see how memory usage grows continiously with time until signal 11 is raised. Unfortunately the debugging symbols are absent, here is the error:
openmx:203644 terminated with signal 11 at PC=8330f7 SP=7fffec8a9928. Backtrace: ...
and the last record in stdout:
Num. of grids overlapping with atom 18 = 232081 Num. of grids overlapping with atom 19 = 232080 Num. of grids overlapping with atom 20 = 232157 .......... ......
I may also post here the input file if you want to try to reproduce it.
Artem
| |