PC GAMESS-related discussion club

Alex Granovsky
gran@classic.chem.msu.su

it is possible that some ordinal numbers of exported functions may have been changed in NT-MPICH version 1.5, making it incompatible with the mpibind.nt-mpich.dll binding file. I tested NT-MPICH 1.5 on our cluster and found it working fine with mpibind.mpich.nt.dll binding dynamic library (which does not use ordinals but references only symbolic names of exported functions).

Best regards,
Alex






On Tue Feb 26 '08, Maurizio Ciofalo wrote
-----------------------------------------
>Hi,
>last days, within our 64 bit Windows PC GAMESS cluster of machines, I tried to upgrade the 1.5.0 NT-MPICH interface (owing to reportedly "many many bugs have been fixed"), which was released in the last weeks of 2007. However, I got bad results due to a fail of actual communication among machines. Typical symptoms are the undertaking by the main node of all the PC GAMESS processes which should have been distributed among the nodes and the fall of the rcluma service in the main node. However, the MPICH lib subdir is correctly addressed n the PATH system variable, the cluster manager service is active in every machine and Windows SP2 Firewall is open. Anyway, everything worked fine with 1.4.0 release. Thus I had to downgrade to the previous MPICH release, eventually.
>Any experience of yours with such a new parallel interface?
>Thank you for the attention
>

[ This message was edited on Sat Mar 1 '08 by the author ]