On Jan 26, 2005, at 9:21 AM, Hubert Feyrer wrote:

> I could interpret this as running "sync" would put the system in a 
> state where I could also suspend it, but obviously that's not the case 
> any longer, with softdep,  and whatever else.

Just running sync is never enough, because the other applications could 
still be writing like gangbusters.

What really needs to happen is for the *kernel* to suspend writing to 
all file systems (which would also wait for all of the I/O to complete) 
right before making the transition to "asleep" state.

-- thorpej