what happened to the nifty ioprio cache hinting stuff?

what happened to the nifty ioprio cache hinting stuff?

[Nix]

I found the ioprio cache hinting quite exciting (I know, I'm weird):
it won't help with writeback or e.g. NFS, but it will certainly let me
do things like run backups without wearing my SSD out.

This was in ewheeler's for-4.10-block-bcache-updates branch, but it
didn't get into 4.10 or 4.11, even while other commits on the same
branch did :( is it dead? (I mean, it still seems to *apply* to 4.11...)

-- 
NULL && (void)

Re: what happened to the nifty ioprio cache hinting stuff?

[Eric Wheeler]

Hi Jens,

It looks like there are still people interested (excited, even!) in the 
sysfs entries for ioprio cache hinting in bcache.  

If I rebase on for-4.12 or for-next on git.kernel.dk, would you pull it in?

There are a handful of stable patches on the bcache list that I would roll 
up in the same pull.  Please let me know what you would prefer and what I 
can do to make it easy for you.

Thank you for your help!

--
Eric Wheeler

On Tue, 2 May 2017, Nix wrote:
> I found the ioprio cache hinting quite exciting (I know, I'm weird):
> it won't help with writeback or e.g. NFS, but it will certainly let me
> do things like run backups without wearing my SSD out.
> 
> This was in ewheeler's for-4.10-block-bcache-updates branch, but it
> didn't get into 4.10 or 4.11, even while other commits on the same
> branch did :( is it dead? (I mean, it still seems to *apply* to 4.11...)
> 
> -- 
> NULL && (void)
> --
> To unsubscribe from this list: send the line "unsubscribe linux-bcache" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
> 

Re: what happened to the nifty ioprio cache hinting stuff?

[Nix]

On 2 May 2017, Eric Wheeler outgrape:

> Hi Jens,
>
> It looks like there are still people interested (excited, even!) in the 
> sysfs entries for ioprio cache hinting in bcache.  

As an aside, I expect to use it for all sorts of things (like backups:
bup likes to reread all data sometimes, and I don't want to blow the
bcache when it does), but in particular I plan to use it for things I
know will be using large volumes of data for the last time before
destruction -- things like manual and automated 'git gc', where I know
that preserving the old packfiles in the bcache is a total waste of time
and where the packfiles being read in only to be deleted may be very
large.

(I'm running in writearound mode, so the lack of useful ioprio for
writeback is not a problem for me, and I can cope with the lack of
ioprio for NFS reads, though I vaguely wish there was a way to propagate
some sort of please-don't-cache-this hint across that channel too!)

Re: what happened to the nifty ioprio cache hinting stuff?

[Kai Krakow]

Am Tue, 02 May 2017 11:54:38 +0100
schrieb Nix <nix@esperi.org.uk>:

> I found the ioprio cache hinting quite exciting (I know, I'm weird):
> it won't help with writeback or e.g. NFS, but it will certainly let me
> do things like run backups without wearing my SSD out.

I'm certainly waiting for the patches to be upstreamed. Meanwhile, I
stopped the hassle of applying the patches and just resorted to putting
my backups partition into "write-around" mode. It still gets big
improvements in performance and is nice to wear leveling.

> This was in ewheeler's for-4.10-block-bcache-updates branch, but it
> didn't get into 4.10 or 4.11, even while other commits on the same
> branch did :( is it dead? (I mean, it still seems to *apply* to
> 4.11...)

Last time I looked at bitbucket (the only place I found it) to pull the
patches via http, it was horribly slow and even complained that the
repository is too big to be browsed or generate patch sets.

Would it be possible to mirror it somewhere else? Bitbucket is really
slow for me, direct git-clone didn't even work.

I was just trying to pull the branch and rebase it ontop of current
kernels myself because I couldn't apply these patches. But I eventually
gave up due to bitbuckets superior *cough* performance... :-(

Also, I cannot find it on git.kernel.org which would've been an
alternative.

Apparently, Eric's user name seems to be already taken on Github -
maybe that's why it isn't there... (or I cannot find it) ;-)


-- 
Regards,
Kai

Replies to list-only preferred.

Re: what happened to the nifty ioprio cache hinting stuff?

[Nix]

On 2 May 2017, Kai Krakow said:

> Am Tue, 02 May 2017 11:54:38 +0100
> schrieb Nix <nix@esperi.org.uk>:
>
>> I found the ioprio cache hinting quite exciting (I know, I'm weird):
>> it won't help with writeback or e.g. NFS, but it will certainly let me
>> do things like run backups without wearing my SSD out.
>
> I'm certainly waiting for the patches to be upstreamed. Meanwhile, I
> stopped the hassle of applying the patches and just resorted to putting
> my backups partition into "write-around" mode. It still gets big
> improvements in performance and is nice to wear leveling.

I'm running in writearound mode regardless, since my bcache is atop an
md array, and that array is journalled on the same SSD (for powerfail
reliability and because in write-back mode it greatly speeds up writes),
and I don't want to burn the SSD out *that* badly. It still speeds
things up hugely :)

>> This was in ewheeler's for-4.10-block-bcache-updates branch, but it
>> didn't get into 4.10 or 4.11, even while other commits on the same
>> branch did :( is it dead? (I mean, it still seems to *apply* to
>> 4.11...)
>
> Last time I looked at bitbucket (the only place I found it) to pull the
> patches via http, it was horribly slow and even complained that the
> repository is too big to be browsed or generate patch sets.
>
> Would it be possible to mirror it somewhere else? Bitbucket is really
> slow for me, direct git-clone didn't even work.
>
> I was just trying to pull the branch and rebase it ontop of current
> kernels myself because I couldn't apply these patches. But I eventually
> gave up due to bitbuckets superior *cough* performance... :-(

It is much faster to take an existing kernel working tree and do a

git remote add ewheeler https://bitbucket.org/ewheelerinc/linux.git
git remote update ewheeler

since this only needs to get a few hundred KiB of objects.

Re: what happened to the nifty ioprio cache hinting stuff?

[Kai Krakow]

Am Fri, 05 May 2017 22:48:45 +0100
schrieb Nix <nix@esperi.org.uk>:

> On 2 May 2017, Kai Krakow said:
> 
> > Am Tue, 02 May 2017 11:54:38 +0100
> > schrieb Nix <nix@esperi.org.uk>:
> >  
> >> I found the ioprio cache hinting quite exciting (I know, I'm
> >> weird): it won't help with writeback or e.g. NFS, but it will
> >> certainly let me do things like run backups without wearing my SSD
> >> out.  
> >
> > I'm certainly waiting for the patches to be upstreamed. Meanwhile, I
> > stopped the hassle of applying the patches and just resorted to
> > putting my backups partition into "write-around" mode. It still
> > gets big improvements in performance and is nice to wear leveling.  
> 
> I'm running in writearound mode regardless, since my bcache is atop an
> md array, and that array is journalled on the same SSD (for powerfail
> reliability and because in write-back mode it greatly speeds up
> writes), and I don't want to burn the SSD out *that* badly. It still
> speeds things up hugely :)
> 
> >> This was in ewheeler's for-4.10-block-bcache-updates branch, but it
> >> didn't get into 4.10 or 4.11, even while other commits on the same
> >> branch did :( is it dead? (I mean, it still seems to *apply* to
> >> 4.11...)  
> >
> > Last time I looked at bitbucket (the only place I found it) to pull
> > the patches via http, it was horribly slow and even complained that
> > the repository is too big to be browsed or generate patch sets.
> >
> > Would it be possible to mirror it somewhere else? Bitbucket is
> > really slow for me, direct git-clone didn't even work.
> >
> > I was just trying to pull the branch and rebase it ontop of current
> > kernels myself because I couldn't apply these patches. But I
> > eventually gave up due to bitbuckets superior *cough*
> > performance... :-(  
> 
> It is much faster to take an existing kernel working tree and do a
> 
> git remote add ewheeler https://bitbucket.org/ewheelerinc/linux.git
> git remote update ewheeler
> 
> since this only needs to get a few hundred KiB of objects.

Yes, that worked... While I've been at that, I also rebased the bfq
patches to get the latest bfq version for ck-sources.

Although it took about a minute to connect to bitbucket - but pulling
the data itself was fast then, matter of eye blinks.

-- 
Regards,
Kai

Replies to list-only preferred.