[dm-devel] [PATCH V3 0/5] dm-rq: improve sequential I/O performance

[Mike Snitzer]

On Thu, Jan 11 2018 at  6:27pm -0500,
Bart Van Assche <Bart.VanAssche at wdc.com> wrote:

> On Thu, 2018-01-11 at 17:58 -0500, Mike Snitzer wrote:
> > The changes are pretty easy to review.  This notion that these changes
> > are problematic rings very hollow given your lack of actual numbers (or
> > some other concerning observation rooted in testing fact) to back up
> > your position.
> 
> It's not my job to run the multi-LUN test. That's the job of the people who
> want these patches upstream. Since I asked for these test results for the first
> time several months ago I'm surprised that nobody has run these tests yet.

I've reasoned through a few different ways to respond to this.  Fact is
you're not giving me much to work with.  AFAIK you _are_ charted with
supporting the types of storage configs that you've requested
performance results from.

Your dm-rq.c commit 6077c2d706097c0 ("dm rq: Avoid that request
processing stalls sporadically") silently went in through Jens:
https://www.redhat.com/archives/dm-devel/2017-April/msg00157.html
Not sure why that happened to begin with honestly.

But at the end of that post I meant to say:
"If this dm-mq specific commit is justified the case certainly is
_not_ spelled out in the commit header."

Anyway, I've split this contentious removal of dm_mq_queue_rq's
blk_mq_delay_run_hw_queue(hctx, 100/*ms*/) into a separate patch; but
at this point I'm still inclined to accept it for 4.16.

I'll hopefully look closer at understanding the need for commit
6077c2d706097c0 tomorrow.

In the meantime, I'd _really_ appreciate it if you'd give the rest of
the changes Ming has proposed in this patchset a much more open mind!

Thanks,
Mike