Hi, Steve! Thanks for this suggestion. I tried various "cutoff" parameters in the diagnostic bondswap command, and it still crashes in the same manner "p4_error: interrupt SIGSEGV: 11 .... Broken Pipe", but the time of the crash depends on the cap. I found quite sharp, reproducible boundaries, which I indicate. I also tried increasing the cutoff. All results are for two processor DBH, for the same system as in my first email.
Cap Crash on Step
1.4 209
1.35 719
1.3 194
1.25 194
1.209 194
1.208 319
1.2 319
1.195 159
1.19 159
1.18 44
1.15 44
1.14 4 (not a typo, really step 4)
1.10 4
All these values run fine over 1000 steps with bondswapping deactivated.
So, it seems there is some problem which occurs only on parallel runs and is highly sensitive to the "cutoff" argument of the diagnostic-bondswap command.
If that doesn't ring a bell, just let me know and I'll go ahead and find a small system that crashes similarly and do the usual things to try to localize what causes the crash; code in some additional warnings, do many more position dumps, etc. (or just send it the script & system to you, if you wish)
Thanks!
Rob