> >	1) change `/sbin/update' to let the time-between-syncs be a
> >	   configurable parameter.  On laptops, you have your own
> >	   UPS, so letting users change the period to 5 or 10 minutes
> >	   instead of 30 seconds would be reasonable.
> 
> Ideally you wouldn't want to update at all.  Since all the memory is
> battery backed you could consider this "stable storage."

No, not exactly, as it gets clobbered across reboots/system
crashes/battery failures.

> >	2) change FFS to keep "fuzzy" access times.
> >	   the proposal is, when updating the atime field, to not
> >	   change it (and not dirty the inode/bufcache page) if the
> >	   timestamp is within <fuzz> of the current time.
> 
> Better: mark the inode as "changed and needs updateting at a
> convenient point".

No, actually, that's closer to how it's done now.., it's just that the
"convenient point" is when sync() is called.

There are three sorts of changes to disk pages:
	1) "important" metadata changes (creating/deleting files/directories)
	2) user data pages
	3) atime updates

I'm absolutely not willing to compromise the integrity of #1 or #2,
but I see there as being some room for optimization in how we handle
#3.  #2 is the *real* mission of the file system; #1 is a means
towards that goal, and #3 is overhead..

> Of course, this isn't i386-specific at all.  Should the thread be
> moved to tech-kern or where?

Perhaps...

					- Bill