>Doing this in general breaks UNIX security as we know it, so it
>*definitely* would have to be a filesystem mount option.
The atime patch is implemented as a mount option (no_atime)
The inode patch seems to be also, but its major intended use is to read
files which are globally readable anyway. In most situations, the news
partition (whether a disk or an md array) is on its lonesome anyway, as
that's the only way to get enough speed out of them.
FWIW, my news machine is averaging 5 article writes per second and about 5
times that figure for reads. The atime patch led to a 400% speedup in UUCP
newsbatching on my system, so the 30% being bandied around can be improved
on in some specialised circumstances (presorted batchlist, enough ram to
buffer/cache things, etc)
AB
-- If you go down to the woods today you're sure for a big surpise: Teddies in leather who bang heads together, with black liner round their eyes.