RE: ext4 dbench performance with CONFIG_PREEMPT_RT

[john stultz <johnstul@us.ibm.com>]

On Fri, 2010-04-09 at 17:48 -0600, Chen, Tim C wrote:
> 
> >tytso@mit.edu wrote 
> >
> >Yeah, I'm very much aware of that.  What worries me is that locking
> >problems in the jbd2 layer could be very hard to debug, so we need to
> >make sure we have some really good testing as we make any changes.
> >
> >Not taking the j_state_lock spinlock in jbd2_stop_lock() was relatively
> >easy to prove to be safe, but I'm really worried about
> >start_this_handle() the locking around that is going to be subtle, and
> >it's not just the specific fields in the transaction and journal
> >handle.
> >
> >And even with the jbd2_stop_lock() change, I'd really prefer some
> >pretty exhaustive testing, including power fail testing, just to make
> >sure we're in practice when/if we make more subtle or more invasive
> >changes to the jbd2 layer...
> >
> >So I'm mot waving the red flag, but the yellow flag (as they would say
> >in auto racing circles).
> >
> 
> Your patch did remove the contention on the j_state_lock for dbench
> in my testing with 64 threads.  The contention point now
> moves dcache_lock, which is also another tricky bottleneck.

Nick Piggin's vfs scalability patches takes care of the dcache_lock
contention. I'm actually using them with the -rt patch in my testing
here.


> In our other testing with FFSB that creates/rename/remove a lot of directories,
> we found that journal->j_revoke_lock was also heavily contended.

Yep. This also shows up in my -rt patch testing with Ted's patch.

thanks
-john