On Mon, Sep 03, 2001 at 11:51:34PM +1000, matthew green wrote:
> 
>    On Mon, Sep 03, 2001 at 11:39:25PM +1000, matthew green wrote:
>    > too much work?  it seems to be too little.  at least at unmount time,
>    > from the softdep umounts listed:
>    
>    I mean "too many I/O operations"..
> 
> 
> from my observations, this isn't the case.  when i've had softdep & rm
> take *forever* on me (ie, the 'df' output takes minutes to actually stop),
> i've run 'vmstat 1' and often the disks are completely idle for 10's of
> seconds at time.

well, disks sitting idle for seconds when there's work waiting to be done
is definitely a problem.  no one had mentioned that before,
and I haven't noticed this on my machines.  this is the first thing
to investigate.


> chuq has given me a test to run next time i'm seeing this, to drop to
> ddb after the rm(1) completes and see how many pending softdeps there
> are.  i think he may believe that there are far far too many.  that i've
> had my kmem_map explode during one of these episodes tends to lead
> credence to this idea...

actually I think there probably aren't way too many, but I wanted to
rule that out so we don't have to think about it anymore.  the next
thing would have been to watch the i/o rate and see if anything interesting
showed up there, but from the previous paragraph it seems we've already
found something interesting.

-Chuck