boost::mpi an order of magnitude slower than pure MPI for sending vector of doubles

[ilja.j.honkonen@…]

As detailed in my email to boost users (h ttp://lists.boost.org/boost-users/2015/02/83807.php) it seems to take an order of magnitude longer to send a vector of doubles using boost::mpi than it takes using C bindings of MPI. I couldn't find anything in the documentation about optimizing transfers of standard types so I'd assume they are already optimized but that doesn't seem to be the case. Am I missing something or is there a serious performance issue in boost::mpi when sending (largeish) vectors of standard types?