With Nick's and Peter's help (thanks guys!), I have been able to get MOLPRO running in parallel on my cluster. Now I have a question:
I managed to locate the procgrp file location (for the stout hearted, it's /usr/local/lib/molpro-mpp-Linux-i686-i4-2002.6) and specified my own (which works just fine). The general form of the content of the procgrp file is: user node number_of_processes path_to_executable work_directory In my case, specifically, the contents are (simplified for this discussion) root n01 2 /usr/molprop_2002_6_p4_tcgmsg.exe /s1 root n02 2 /usr/molprop_2002_6_p4_tcgmsg.exe /s1 root n03 2 /usr/molprop_2002_6_p4_tcgmsg.exe /s1 root n04 2 /usr/molprop_2002_6_p4_tcgmsg.exe /s1 Is this functionally the same as: root n01 2 /usr/molprop_2002_6_p4_tcgmsg.exe /s1 root n01 2 /usr/molprop_2002_6_p4_tcgmsg.exe /s2 root n02 2 /usr/molprop_2002_6_p4_tcgmsg.exe /s1 root n02 2 /usr/molprop_2002_6_p4_tcgmsg.exe /s2 root n03 2 /usr/molprop_2002_6_p4_tcgmsg.exe /s1 root n03 2 /usr/molprop_2002_6_p4_tcgmsg.exe /s2 root n04 2 /usr/molprop_2002_6_p4_tcgmsg.exe /s1 root n04 2 /usr/molprop_2002_6_p4_tcgmsg.exe /s2 ? When I run each choice of procgrp, I see no net effect on 'real' time. Under MPICH, with other software, I can specify the second type of procgrp and see a substantial reduction in real time since each processor is treated as a separate instance of that executable. Am I correct in my conclusion that MOLPRO/ Global Arrays does not do the same as MPICH? Does MOLPRO/ Global Arrays take the specification of 2 processes for a node and just turn on SMP processing? Is there a way I can get MOLPRO to provide the same behavior I have seen with MPICH? Any insight that anyone can provide would be appreciated. Also, does anyone know why MOLPRO ignores the work_directoy specified in the procgrp file? I will summarize responses to the list. Jim
