[Help-glpk] Re: parallelism and glpk

[Andrew Makhorin]

> I implemented a simulated annealing, where in the inner cycle I solve
> many transportation problems with glpk.

> Then I parallelized it with MPI, and got great results almost linear
> speedup.

> Now I parallelized it with a distributed shared system, not using
> threads, and somehow glpk its messing with the virtual memory of the DSM
> system, because when I use the virtual memory the implementation
> sometimes dies with segmentation fault, I said sometimes, because
> sometimes ends OK.

> The DSM system I'm using is page based, and it uses threads to manage the
> page requests.

It is difficult to say what precisely happens. Could you provide more
details? In particular, could you determine in which glpk routine the
error occurs? Please note that glpk is not re-entrant and has no
synchronization features, so glpk routines must not be called from
parallel tasks within the same address space.