On Mon, Feb 24, 2003 at 11:09:24AM -0800, Bill Studenmund wrote:
> I think instead you want to mount it write-defered, a state we don't have
> yet. :-)  The difference is that it's ok to be open r/w, but if you start
> to do something that will update metadata, you block until "write-defered"
> is cleared.
> 
> Hmmm.. It might be ok to ignore certain updates in this case, rather than
> defering them. Like you may well want to keep reads working, and just
> throw away the atime change - implicitly set noatime and nodevtime.

There are many little ideas and uses for similar kinds of changes
lurking out there.   One of my pet wishes has to do with spinning
down disks for power-saving in laptops.  I mount noatime, nodevmtime,
sure - but even then the disk spins up sometimes when i really
don't especially need it to right then and there.

In amongst all the various other things people are considering, it
would be nice to have some kinds of writes (atime updates are an
obvious choice) queued at a priority that says "don't spin up the
disk just for this write, wait until you have to spin it up for
something else".

It's really starting to seem that there's enough interconnectedness
between all these various little ideas that it needs someone to
take ownership of some consolidated work to tie them together into
a framework. Any volunteers? :)

--
Dan.