Re: [boost] [Parallel BGL] Property access problem

Subject: Re: [boost] [Parallel BGL] Property access problem
From: Cosimo Calabrese (cosimo.calabrese_at_[hidden])
Date: 2011-04-14 04:36:05

I've the suspect that with 2 processors, the program inserts the edges 
till their vertices belong to the same process. I presume that the first 
process has the vertices A, B, C, D, E, and the second process has F, G, 
H, I, L. The crash happens when I insert the D->G edge, from a process 
to another; actually I cannot insert an edge that involves vertices that 
not belong to the root process, not only "interprocess edges". For 
example, I cannot insert a vertex G->L.

I've also compiled the hohberg_biconnected_components_test, and it works 
fine with 1, 2, 3, 4 processes.

Can someone help me?

Thanks in advance,
Cosimo Calabrese.

Il 13/04/2011 16.22, Cosimo Calabrese ha scritto:
> Hi to all,
>
> I'm working on a WinXP, VS2010, boost 1.46.1, openMPI 1.4.3.
> I've used in the past the BGL lib, and now I would explore the PBGL
> extension. I've writed a simple program (you can find it in attach), but
> I've some problem. When I run it with:
>
> mpirun -np 1 .\test.exe
>
> it works. But if I use 2 or 4 processors the program crash
> sistematically in the following way, on different machines too:
>
> num_vertices: 10
> num_edges: 19
> A-->B : 1
> A-->C : 10
> A-->D : 11
> B-->D : 13
> C-->D : 12
> B-->E : 1
> C-->A : 14
> E-->D : 1
> D-->B : 15
> D-->A : 1
> E-->B : 16
> --------------------------------------------------------------------------
> Sorry! You were supposed to get help about:
> odls-default:could-not-kill
> But I couldn't open the help file:
>
> D:\dev\CommesseAperte\Sviluppo\TELLUSXX_CalcServer\lib\openmpi\installed\share\openmpi\help-odls-default.txt:
> No such file or directory. Sorry!
> --------------------------------------------------------------------------
>
> Every process allocates 1 GB.
>
> The problem disappear only when I eliminate all the property access
> (get() and put()).
>
> Maybe I access to the MPI framework for the first time during the
> property access, and so could be an MPI problem? Am I using the property
> in a wrong way?
>