Firefly and PC GAMESS-related discussion club



Learn how to ask questions correctly


Re^6: Firefly Bug

Alex Granovsky
gran@classic.chem.msu.su


Dear Alex,

By the way, could you please clarify several important points?

First, I did not completely understand whether you are running
Firefly in parallel mode, or in serial just using several working
threads specified via mklnp or np variable in the $system group;
or both?

Second, what is the exact version (the exact filename of the
Firefly's distribution archive) you are using? The point is
that the fully statically linked MPICH version
uses old threading model (i.e., it does not use pthreads) and
indeed internally it uses sysvipc semaphores. All other versions
uses pthreads threading model and do not use semaphores at all -
however, MPICH itself (that is configured to use ch_p4 device)
uses them. Thus, even semi-statically linked MPICH version
indirectly uses semaphores. Versions for all other MPI
implementations may or may not indirectly use semaphores depending
on the details of the particular implementation of MPI. By the way,
do you run cleanipcs after forcibly killing Firefly processes that
uses semaphores and shared memory?

Next, are you using dlb or rather extended dlb (xdlb) model of p2p protocol?


Finally, you may find the following thread of some interest to you:
http://classic.chem.msu.su/cgi-bin/ceilidh.exe/gran/gamess/forum/?C353e1b6dadzG-6933-704+00.htm
In brief, there are serious compatibility issue with MPICH and
TCP/IP stack implementations of some modern Linuxes...

Regards,
Alex Granovsky







On Mon Feb 1 '10 4:15am, alex wrote
-----------------------------------
>Thank you a lot!
>It takes some time to update OS and check if it is cure.
>

[ This message was edited on Tue Feb 2 '10 at 12:57pm by the author ]


[ Previous ] [ Next ] [ Index ]           Tue Feb 2 '10 12:57pm
[ Reply ] [ Edit ] [ Delete ]           This message read 740 times