I recently ran into the problem that a local container on the stack, which was likely to be very small, but its size was still not known at compile-time, consumed a considerable amount of runtime because of allocation/deallocation. e.g. void do(istream &i){ vector<int> vec; vec.reserve(20); while(...){ int tmp; i >> tmp; vec.push_back(tmp); } } so I wrote an allocator that I think is generally useful and could be added e.g. to Boost.Array: void do(int n){ int buffer[20]; stack_allocator<int,20> alloc(buffer); vector<int,stack_allocator<int,20> > vec(alloc); vec.reserve(alloc.max_stack_size()); while(...){ int tmp; i >> tmp; vec.push_back(tmp); } } the vector operates on the stack as long as there is enough space, and moves to the heap afterwards. in most cases there is no dynamic allocation. here's the source: http://www.boostpro.com/vault/index.php?action=downloadfile&filename=stack_allocator.hpp&directory=array& it uses the simple_seq_fit allocation algorithm described here: http://www.boost.org/doc/libs/1_37_0/doc/html/interprocess/allocators_contai...