the recently implemented prd functionality is of great interest to me, since i am studying diffusion processes on metal surfaces. So i decided to play a bit with it.
I downloaded and compiled the recent distribution (timestamp 13Nov) of LAMMPS and started by comparing the logfile output of my machines with the one provided in the example directory. Besides the to be expected differences in the run time a diff to those fles showed no differences in the output. Moreover, all replica MD simulations seem to finalize withour any errors with respect to their logfiles.
However, the last LAMMPS statement is an error message, thrown to the operating system. On a shared memory dualcore Ubuntu 9.10 Linux box with the standard ubuntu mpich installation it is a "Segmentation fault" and on a cluster of linux boxes with an mpich-1 environment it is another message, that is also basically stating that a write process could not be finished. (sorry, i didn't save the output)
I tried to have a look at the cpp sources, but since i am not so familiar with MPI Programming, i though i would post this issue to the list, maybe another one could have a look.