Re: high overhead of functions blkg_*stats_* in bfq

Il giorno 06 nov 2017, alle ore 16:00, Tejun Heo [off-list ref] ha =

scritto:

=20
Hello,
=20
On Sun, Nov 05, 2017 at 07:21:01PM -0700, Jens Axboe wrote:

quoted

It's pointless to give up on this so soon, when no effort has =

apparently

quoted

been dedicated to figuring out what the actual issue is yet. So no, =

no

quoted

patch that will just disable the stats is going to be accepted.
=20
That said, I have no idea who uses these stats. Surely someone can
answer that question. Tejun?

=20
Except for the basic bytes / ios counts, it's all debug fluff, which
should have been hidden behind a debug boot param or go under debugfs.
I'm not sure we can get rid of them at this point for cfq but I don't
see why we'd have them for bfq.
=20

Ok, then I think this is the right time to ask you what I can throw
way for bfq.  According to your documentation, basic non-debug stats
should be:

blkio.time
blkio.time_recursive
blkio.sectors
blkio.io_service_bytes
blkio.io_service_bytes_recursive
blkio.io_serviced
blkio.io_serviced_recursive
blkio.io_service_time
blkio.io_service_time_recursive
blkio.io_wait_time
blkio.io_wait_time_recursive
blkio.io_merged
blkio.io_merged_recursive
blkio.io_queued
blkio.io_queued_recursive

So, I have to keep them in bfq, right?  Or is there something I can
remove for bfq, in your opinion?  Of course, I would be very happy to =
remove stuff!

On the other end, here are the debugging stats I have borrowed
from cfq:

blkio.avg_queue_size
blkio.dequeue
blkio.empty_time
blkio.group_wait_time
blkio.idle_time
blkio.io_wait_time
blkio.io_wait_time_recursive

These will be hidden behind the CONFIG_DEBUG_BLK_CGROUP option by one of =
the patches we are about to submit. Or do you want me to remove these =
debugging stats completely? ok for me, just, if these parameters are =
removed, then I guess the documentation of cgroups-v1 will have to be =
somehow updated once the interface of bfq and cfq is unified.

Thanks,
Paolo

Thanks.
=20
--=20
tejun