It appears that the named allocation functions for the managed memory classes do not place constructed objects at the appropriate alignment for the requested type in all situations. Here is a simple example to illustrate the issue. #include <iostream> #include <boost/interprocess/managed_mapped_file.hpp> // alignment of 16 on i386 and x86_64 struct vector_of_double { long double x; long double y; long double z; }; int main() { namespace bip = boost::interprocess; { std::cout << "Alignment of vector_of_double is " << __alignof__(vector_of_double) << std::endl; bip::managed_mapped_file simple_file_mapping(bip::create_only, "simple_mapping", 65536); vector_of_double * test_vec = simple_file_mapping.construct<vector_of_double>("a named vector allocation")(); std::cout << "Resulting allocation is at address: " << test_vec << std::endl; } bip::remove_file_on_destroy("simple_mapping"); return 0; } On gcc targeting i386, this code prints: Alignment of vector_of_double is 16 Resulting allocation is at address 0x???????c Similar results on x86_64. It appears the resulting address is only aligned on a 4 byte boundary on 32 bit targets and 8 byte on 64 bit targets. Are there workarounds? I have tried explicitly setting the alignment parameter in the mapped region's index parameter to 16, but it did not change the result. I have tested the problem with anonymous allocations, and those do appear to be appropriately aligned. Only named allocations have the issue.