Waiting time is too long
Shawn Yang
shawn.yan... at gmail.com
Thu Mar 17 15:02:14 UTC 2016
Hi all,
I am doing a Mixed mode parallel simulation on cp2k 3.0. The package was
installed by using the install_toolchain script.
Here is part of the performance when running:
-------------------------------------------------------------------------------
-
-
- MESSAGE PASSING PERFORMANCE
-
-
-
-------------------------------------------------------------------------------
ROUTINE CALLS TOT TIME [s] AVE VOLUME
[Bytes] PERFORMANCE [MB/s]
MP_Group 5 0.000
MP_Bcast 57138 75.397
10390. 7.87
MP_Allreduce 113420 2192.750
296. 0.02
MP_Gather 102 0.929
640. 0.07
MP_Sync 63 83.271
MP_Alltoall 113191 2289.461
533721. 26.39
MP_SendRecv 6324 6.835
1283. 1.19
MP_ISendRecv 262260 1.986
51157. 6755.54
MP_Wait 2340964 3375.440
MP_comm_split 9 0.438
MP_ISend 1357992 9.288
34472. 5039.93
MP_IRecv 1358200 2.225
34115. 20822.97
MP_Recv 770 2.647
178840. 52.02
MP_Memory 1400158 0.753
-------------------------------------------------------------------------------
The total time for this calculation is 3 hours---10800s, so that means
"MP_WAIT" takes nearly 1 hour. Anyone knows is it normal?
Following is the libs I used when running:
CP2K| version string: CP2K
version 3.0
CP2K| source code revision number:
svn:16458
CP2K| cp2kflags: libint fftw3 libxc pexsi elpa3 parallel mpi3 scalapack
quip sm
CP2K| m_dnn smm libderiv_max_am1=5 libint_max_am=6
CP2K| is freely available from
https://www.cp2k.org/
CP2K| Program compiled at Wed Mar 16 01:20:41 GMT
2016
CP2K| Program compiled on
login1
CP2K| Program compiled for
local
Please give me some advise, the calculation is too slow, there must be
something wrong. I've tried same input file and cores in ARCHER, it's 5
time faster.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20160317/7e75c4c2/attachment.htm>
More information about the CP2K-user
mailing list