On Wed, Jul 12, 2006 at 03:07:41PM +0400, Bulat Ziganshin wrote:
Using memory-mapped files provides slight speed improvement because it excludes the need to move data between OS buffer and application buffer. Moreover, this may scale better in multi-threading environments without need to implement special async I/O, although i don't checked this assumption.
There's also the potentially quite significant advantage that with mmapped files you can open up files much larger than the physical memory size. Or if you open smaller files, but the system begins to swap, mmaped files don't have to be swapped to a swap file, which makes things more efficient in general. In my opinion this is a much more important reason to use mmap than avoiding a memory copy from kernel space to user space. -- David Roundy