[PATCH] UBI: Add volume read and write statistics

[derosier@xxxxxxxxx (Steve deRosier)]

On Tue, Jul 17, 2018 at 8:01 AM Per F?rlin <per.forlin at axis.com> wrote:
>
> Thanks Richard for your feedback,
>
> > ________________________________________
> > From: Richard Weinberger <richard at nod.at>
> > Sent: Tuesday, July 17, 2018 4:34 PM
> > To: Per F?rlin
> > Cc: linux-mtd at lists.infradead.org; Artem Bityutskiy
> > Subject: Re: [PATCH] UBI: Add volume read and write statistics
> >
> > Per,
> >
> > Am Dienstag, 17. Juli 2018, 14:08:51 CEST schrieb Per F?rlin:
> > > > To: Per F?rlin
> > > > Cc: linux-mtd at lists.infradead.org; Artem Bityutskiy; Per F?rlin
> > > > Subject: Re: [PATCH] UBI: Add volume read and write statistics
> > > >
> > > > Am Dienstag, 17. Juli 2018, 12:30:19 CEST schrieb Per Forlin:
> > > > > Simple read and write statistics.
> > > > > * Bytes read
> > > > > * Bytes written
> > > > > * Number of reads
> > > > > * Number of writes
> > > > >
> > > > > This is useful to find out how the storage is being utilized.
> > > > > For block devices this already exists via /proc/diskstats.
> > > > > The intention of this patch is to add similar stats
> > > > > for UBI as well.
> > > >
> > > > Why on UBI level and not MTD?
> > > In my case I wanted to evaluate the performance per volume. I have one MTD
> > > device with several UBI volumes. It would be sufficient to have it on
> > > an MTD level to see the overall storage usage.
> > > This would still be very helpful for me.
> > >
> > > Having it on an MTD level is of course more general.
> > > I wouldn't mind changing the patch to add the stats in mtdcore
> > > for mtd_read() and mtd_write()
> > >
> > > In case of MTD block devices the stats will be somewhat redundant with
> > > /proc/diskstats.
> > >
> > > Do you think I should update the patch to add MTD stats instead?
> >
> > While having a cup of coffee I thought more about this.
> > Actually both, MTD and UBI makes sense.
> > The most important issue is that you integrate it with the existing diskstats.
> > So instead of having our own interface feeding MTD/UBI stats into diskstats
> > would be nice. Did you look into that? I'm not sure how much work this would be.
> > That way users can use existing tools such as iostat...
> I actually started out looking for the information under diskstats,
> then I learned it's only for block devices. I took a quick glance at
> it before I went for the sys implementation instead. diskstats is
> separated from the MTD and UBI stuff and I don't know if one can make a
> connection to MTD/UBI somehow. I will take a closer look at this.

Perhaps it was "only for block devices" because no one ever
implemented the necessary hooks in MTD or UBI?  I don't know the
history, nor the information you found, just making a stab in the
dark.

If UBI and/or MTD can provide the statistics that diskstats needs in a
interpretation that makes sense, why not go that way?

- Steve