[MOAB-dev] Parallel output writes odd sets

[Grindeanu, Iulian R.]

Hi Anton,
most of those extra sets are the shared sets between parts. 
also, there are some sets that comprise all entities on a task

some of them are used during ghosting

You should have all the original sets, too.

If it becomes a problem for you, please raise an issue on bitbucket, and we may offer an option to not write them 
(maybe there is one now, but I do not know it)

Iulian

________________________________________
From: moab-dev-bounces at mcs.anl.gov [moab-dev-bounces at mcs.anl.gov] on behalf of Kanaev A.A. [kanaev at ibrae.ac.ru]
Sent: Thursday, October 30, 2014 11:00 AM
To: moab-dev at mcs.anl.gov
Subject: [MOAB-dev] Parallel output writes odd sets

Hello MOAB-dev,

I have a simple program that reads then writes h5m file in parallel
The input file is brick1x1x2p2.h5m
The output file is out.h5m

Output of mbsize -ll brick1x1x2p2.h5m shows there're 29 EntitySets : 1-27 are geometry sets, 28 and 29 are partition sets made by mbpart

I run my program in parallel :  mpiexec.mpich2 -n 2 ./restart brick1x1x2p2.h5m
Output of mbsize -ll out.h5m shows there're 33 EntitySets

Loading another larger mesh prepartitioned for 16 processors,  there're +4 sets when running on 2 cores, +16 when running on 4 cores,  +120 sets when running on 16 cores
(after removing PARALLEL_RESOLVE_SHARED_ENTS option for load_file, there's +97 set for 16 processors)

What are those extra EntitySets?

cpp code + h5m file attached

Thanks
Anton