On Mon, Jan 22, 2007 at 05:13:44PM -0600, John Stanton wrote: > >Of course. The point is that 32-bit memory models will impact the > >design of the this pager. Either only support 64-bit memory models, > >window your memory mappings, or only support small databases. > > > Or be adaptive and sense the appropriate model then bind the appropriate > low level API.
It wouldn't be a matter of which low-level API to use but which implementation strategy to use. If you can assume that you can memory map an entire file holding the db then the implementation of the pager could be simpler than if you can't make that assumption. One could, and would use mmap(2) in both cases (in Unix-y environments anyways). Building an internal API that obtains a memory mapped address for some region of some file that is potentially larger than the VM address speace is certainly possible, but when you add in multi-threading it gets painful. ----------------------------------------------------------------------------- To unsubscribe, send email to [EMAIL PROTECTED] -----------------------------------------------------------------------------