I first raised the issue of mmap in SQLite 2 years ago. http://www.mailinglistarchive.com/html/sqlite-dev@sqlite.org/2011-09/msg00023.html http://www.mailinglis

I first raised the issue of mmap in SQLite 2 years ago. http://www.mailinglistarchive.com/html/sqlite-dev@sqlite.org/2011-09/msg00023.html http://www.mailinglistarchive.com/html/sqlite-dev@sqlite.org... Richard started playing with it a couple months later http://www.mailinglistarchive.com/html/sqlite-dev@sqlite.org/2011-11/msg00014.html http://www.mailinglistarchive.com/html/sqlite-dev@sqlite.org... but didn't get any performance benefit at the time because he was Doing It Wrong(tm). SQLite does a lot of excessive work; this work is justified if you're using it on an embedded processor that doesn't have a full virtual-memory based OS underneath. But on any other platform it's a waste of CPU/power/memory/time. And it's been decades since the majority of embedded applications ran on bare metal without some kind of OS. This new mmap feature in SQLite promises "up to 2x" performance improvement, while LMDB in SQLite still yields 20x improvement over stock SQLite.