[mvapich-discuss] mvapich2 warning: Rndv Receiver is receiving less than as expected

[Bernd Kallies]

Hello,

I'd like to report again about a long-lasting problem.
I posted a problem report with the subject like the one of this email
the first time in June 2009. I observed things like that with quantum
chemistry applications like cp2k or cpmd with larger problems.

I provided example inputs to the mvapich2 dev team, which showed this
problem reproducible on our machines. When new mvapich2 versions were
released, I repeated my tests, always without success.

The last time a similar problem was reported by Aaron Knister to the
mvapich2-discuss mailing list on June 29, 2010 (with some v1.4 and
gromacs).

I tried again with mvapich2-1.5.0, and the problem still remains. With
my CPMD example input I provided some months ago to the mvapich2 dev
team, I get aborts for all kind of MV2_RNDV_PROTOCOL settings.
Even with MV2_RNDV_PROTOCOL=RGET I get messages like

"Fatal error in MPI_Alltoall: MPIDI_CH3U_Post_data_receive_found(441):
Message from rank 78 and tag 9 truncated; 466944 bytes received but
buffer size is 118784"

Note that this behaviour prevents us from using mvapich2 for all our
chemistry codes (except VASP, which did not show this error so far).

Is there somebody working on this problem, or is it believed to be
solved somehow?

Sincerely BK

-- 
Dr. Bernd Kallies
Konrad-Zuse-Zentrum für Informationstechnik Berlin
Takustr. 7
14195 Berlin
Tel: +49-30-84185-270
Fax: +49-30-84185-311
e-mail: kallies at zib.de