We try to calculate various (simple) problems with Linux on different PC's (specified Red Hat release and Absoft Compiler). The calculation of the exact same problem aborts randomly on different iteration steps. We don't have this problem on our SGI's (Octane,IRIX). Also the STAR support runs the same problem on linux without failures. We suggest a hardware problem (CPU, board, ram, floating point ...). Does anybody know anything about this, or had the same problems?

Which RedHat version are you running? We have experienced problems with the original RedHat 6.2 kernel. There is a bug in the Linux kernel versions 2.2.13-0.13, 2.2.14-5.0 (which is installed by RedHat 6.2), and 2.2.14-6.1 on x86. Details can be found at:

http://www.redhat.com/support/errata...000013-01.html

Under heavy loads, random data corruption by the operating system can occur. It appears that large memory CFD runs will load a system heavily enough to produce this problem. The symptoms showed up as NaN's in the solver output in our applications because the data files were corrupted.

Just install a newer kernel such as 2.2.14-12 or later. That fixed the problem for us.

I had a similar problem on an NT box running STAR. It turned out that my processor was an OEM product that was over heating, even though it was not overclocked.

I solved the problem by replacing the CPU with a retail boxed chip with genuine INTEL heat sink and fan.