If you do a random seek benchmark, readahead might (depending on the
buffer size and the file's size) make performance worse, but for
sequential access, it would give good performance.
Forgive me not having looked into the sources, but my estimates are
that with more than 50% probability these functions are not there,
but seem to be useful for performance applications like benchmarks or
databases.
That just produced an other idea: What about a WRITE_ONLY hint that
frees the associated buffers immediately after the data is written to
disk (thinking of syslog and other logs, but especially for database
redo logs).
Awaiting comments...
Ulrich