Segment violation while running Vdbench at high IOPS workloads on aura2.
Vdbench is written in 99+% java, with the ultimate open/close/read/write is done in C code (JNI). When running at high iops Vdbench will frequently (9 out of 10 times) abort with segment violations, though some times infrequently this will also happen at low iops. Though the current violation is 'identified' is occurring in Java_Vdb_Native_multi_1io, during a previous test cycle about a month ago I also saw them in _malloc_unlocked, realfree, strcmp, t_splay .
During the previous test cycle when I had access to the hardware I added extra debugging code to Vdbench to determine if I could determine when an possible memory corruption occurred. I was not successful. One scenario however was interesting which I'll try to describe here in pseudo code:

int x = 5;
if (x != 5)
  printf("Current value of x: %d\n", x);
resulting in this output: Current value of x: 5

This scenario happened several times.

I alas currently do not have access to the hardware so I can not repeat these tests.
I am including two sets of data of problems we encountered over the weekend.
- File f40_1fmod00128k_1file_16c_t4_short001_seg_v.zip : vdbench output directory. Look at the bottom of file localhost-3.stdout.html.
- File hs_err_pid13949.log created during this problem.
- File 'core' core dump for this problem.

A second occurrence (interesting: this is the original aura!!! F20)
- File f20_1fmod008k_1file_16c_t4_short001_seg_v.zip, see the bottom of file localhost-3.stdout.html
- File hs_err_pid12223.log
- File core2.