a friend of mine and me have written a sql engine, and we used
our own cache system.
We think its better because we have a better control on the way the
cache is managed. (e.g. we can cache the disk page but the tuple too)
its possible to have some priorities in the cache for different tables.
and even if linux has a good cache, its not true on every unix. so it's more
portable for us not to trust the os cache.
moreover, we avoid the overhead due to the gate of the read() syscall.
and i havn't try to use mmap. so i don't know the result and if its a good
idea in my case.
-- Jerome Etienne <jetienne@ifhamy.insa-lyon.fr> aim: be a pebble in a torrent.