>Agreed. But not at the driver level. This needs to be done at a higther level,
>so that this can benefit for SCSI disks (and possibly others ?) too.

Oh, yeah, sorry, I of course was sprinkling a large dose of "generic
multilayer pixie dust" in my approach :-)

>I have some idea for moving update to a kernel thread, which could have
>an intelligent buffer sync policie (no need to wakeup the disk when the only
>change to flush on disk is an atime update, for example). This kernel
>thread could also do disk PM.
>Unfortunably I don't have the time to work on this now.

Same here :-/

--Ken