Re: [rfc][patch 2.6.18-rc7] block: explicit plugging

[Nick Piggin <npiggin@suse.de>]

On Fri, Oct 06, 2006 at 01:57:31PM +0200, Jens Axboe wrote:
> On Sat, Sep 16 2006, Nick Piggin wrote:
> >   
> > On a parallel tiobench benchmark, of the 800 000 calls to __make_request
> > performed, this patch avoids 490 000 (62%) of queue_lock aquisitions by
> > early merging on the private plugged list.
> 
> Nick, this looks pretty good in general from the vm side, and the
> concept is nice for reduced lock bouncing. I've merged this for more
> testing in a 'plug' branch in the block repo.

Thanks, glad you like it :)

I had a browse through your git branch and it looks pretty sane.
The queue delay looks like a nice elegant fix for the stuff I butchered
out. I didn't think it would take a huge amount of fixing, but I'm
glad you did it, because I know very little about SCSI :P


> > Testing and development is in early stages yet. In particular, the lack of
> > a timer based unplug kick probably breaks some block device drivers in
> > funny ways (though works here for me with SCSI and UML so far). Also needs
> > much wider testing.
> 
> Your SCSI changes are pretty broken, I've fixed them up. We need some
> way of asking the block layer to back off and rerun is sometime soon,
> which is what the plugging does in that case. I've introduced a new
> mechanism for that.
> 
> Changes:
> 
>     - Don't invoke ->request_fn() in blk_queue_invalidate_tags
> 
>     - Fixup all filesystems for block_sync_page()
> 
>     - Add blk_delay_queue() to handle the old plugging-on-shortage
>       usage.
> 
>     - Unconditionally run replug_current_nested() in ioschedule()
> 
>     - Merge to current git tree.

All looks good to me... I don't know about namespace though: do you
think prepending a blk_ or block_ to the plug operations would be nicer?

> I'll try to do some serious testing on this soon. It would also be nice
> to retain the plugging information for blktrace, even if it isn't per
> queue anymore. Hmmm.

I guess you still merge against a particular queue, because you'll
flush the private list when submitting to a different queue. However
trying to combine the stats when you don't hold the queue lock might
be interesting? I guess you don't want to reintroduce any cacheline
bouncing if you can help it.

I will be very interested to know what happens to performance in IO
critical applications.

Thanks,
Nick