[linux-lvm] How does LVM-Cache work?

[linux-lvm] How does LVM-Cache work?

[Leo Antonio]

[-- Attachment #1: Type: text/plain, Size: 736 bytes --]

Hello, everyone.

I'm willing to do research about second-level cache using SSDs and I think
that the best way to do this is using LVM/Device Mapper, because other
implementations are based on FUSE, virtualization layers, and others.

What I'm trying to do is to develop a subsystem that will put hot
blocks/PEs on SSDs-Cache, as well as group related blocks/PEs on cache
automatically. Not simply put on SSD the most  used files or directories.

The problem is that I could not find any documentation about DM or LVM to
help my research and development.

Could you point me to some documentation or what's the best way to
accomplish this?

Thanks!

Leonardo Antônio dos Santos
Federal University of Paraná (Brazil)

[-- Attachment #2: Type: text/html, Size: 900 bytes --]

Re: [linux-lvm] How does LVM-Cache work?

[Zdenek Kabelac]

Dne 26.2.2013 19:44, Leo Antonio napsal(a):
> Hello, everyone.
>
> I'm willing to do research about second-level cache using SSDs and I think
> that the best way to do this is using LVM/Device Mapper, because other
> implementations are based on FUSE, virtualization layers, and others.
>
> What I'm trying to do is to develop a subsystem that will put hot blocks/PEs
> on SSDs-Cache, as well as group related blocks/PEs on cache automatically. Not
> simply put on SSD the most  used files or directories.
>
> The problem is that I could not find any documentation about DM or LVM to help
> my research and development.
>
> Could you point me to some documentation or what's the best way to accomplish
> this?

Look for all-caches branch for dm-cache target:
https://github.com/jthornber/linux-2.6/branches

My assumption is - it's most probably already developed and on its way for 
upstream inclusion.

Zdenek

Re: [linux-lvm] How does LVM-Cache work?

[thornber]

On Tue, Feb 26, 2013 at 03:44:43PM -0300, Leo Antonio wrote:
> Hello, everyone.
> 
> I'm willing to do research about second-level cache using SSDs and I think
> that the best way to do this is using LVM/Device Mapper, because other
> implementations are based on FUSE, virtualization layers, and others.
> 
> What I'm trying to do is to develop a subsystem that will put hot
> blocks/PEs on SSDs-Cache, as well as group related blocks/PEs on cache
> automatically. Not simply put on SSD the most  used files or directories.
> 
> The problem is that I could not find any documentation about DM or LVM to
> help my research and development.
> 
> Could you point me to some documentation or what's the best way to
> accomplish this?

See my dm-cache target in this branch of my tree:

https://github.com/jthornber/linux-2.6/tree/thin-dev

https://github.com/jthornber/linux-2.6/blob/thin-dev/drivers/md/dm-cache-target.c

There's still a lot of work to do on the policy plug-ins, so if you
like to get involved that would be great.

- Joe

Re: [linux-lvm] How does LVM-Cache work?

[Leo Antonio]

[-- Attachment #1: Type: text/plain, Size: 1646 bytes --]

Thanks for answers.


I take a look in dm-cache-target, I really dont knew this project.

Joe, what is the best way to continue this discussion?

Do you have some documentation besides what is into kernel source, like
Papers, Architecture Projects etc?
What is the best way to start to understand this?


Thanks!


2013/2/27 <thornber@redhat.com>

> On Tue, Feb 26, 2013 at 03:44:43PM -0300, Leo Antonio wrote:
> > Hello, everyone.
> >
> > I'm willing to do research about second-level cache using SSDs and I
> think
> > that the best way to do this is using LVM/Device Mapper, because other
> > implementations are based on FUSE, virtualization layers, and others.
> >
> > What I'm trying to do is to develop a subsystem that will put hot
> > blocks/PEs on SSDs-Cache, as well as group related blocks/PEs on cache
> > automatically. Not simply put on SSD the most  used files or directories.
> >
> > The problem is that I could not find any documentation about DM or LVM to
> > help my research and development.
> >
> > Could you point me to some documentation or what's the best way to
> > accomplish this?
>
> See my dm-cache target in this branch of my tree:
>
> https://github.com/jthornber/linux-2.6/tree/thin-dev
>
>
> https://github.com/jthornber/linux-2.6/blob/thin-dev/drivers/md/dm-cache-target.c
>
> There's still a lot of work to do on the policy plug-ins, so if you
> like to get involved that would be great.
>
> - Joe
>
> _______________________________________________
> linux-lvm mailing list
> linux-lvm@redhat.com
> https://www.redhat.com/mailman/listinfo/linux-lvm
> read the LVM HOW-TO at http://tldp.org/HOWTO/LVM-HOWTO/
>

[-- Attachment #2: Type: text/html, Size: 2606 bytes --]

Re: [linux-lvm] How does LVM-Cache work?

[Greg Zapp]

I see reference of another dm-cache from ming zhao as far back as
2007.  Found a link here https://github.com/mingzhao/dm-cache.  Does
anybody know the history behind this stuff?  I would have expected one
of these solutions to be pushed upstream by now considering how much
interest is around tiered storage..

On Thu, Feb 28, 2013 at 7:30 AM, Leo Antonio <heiligerstein@gmail.com> wrote:
> Thanks for answers.
>
>
> I take a look in dm-cache-target, I really dont knew this project.
>
> Joe, what is the best way to continue this discussion?
>
> Do you have some documentation besides what is into kernel source, like
> Papers, Architecture Projects etc?
> What is the best way to start to understand this?
>
>
> Thanks!
>
>
> 2013/2/27 <thornber@redhat.com>
>
>> On Tue, Feb 26, 2013 at 03:44:43PM -0300, Leo Antonio wrote:
>> > Hello, everyone.
>> >
>> > I'm willing to do research about second-level cache using SSDs and I
>> > think
>> > that the best way to do this is using LVM/Device Mapper, because other
>> > implementations are based on FUSE, virtualization layers, and others.
>> >
>> > What I'm trying to do is to develop a subsystem that will put hot
>> > blocks/PEs on SSDs-Cache, as well as group related blocks/PEs on cache
>> > automatically. Not simply put on SSD the most  used files or
>> > directories.
>> >
>> > The problem is that I could not find any documentation about DM or LVM
>> > to
>> > help my research and development.
>> >
>> > Could you point me to some documentation or what's the best way to
>> > accomplish this?
>>
>> See my dm-cache target in this branch of my tree:
>>
>> https://github.com/jthornber/linux-2.6/tree/thin-dev
>>
>>
>> https://github.com/jthornber/linux-2.6/blob/thin-dev/drivers/md/dm-cache-target.c
>>
>> There's still a lot of work to do on the policy plug-ins, so if you
>> like to get involved that would be great.
>>
>> - Joe
>>
>> _______________________________________________
>> linux-lvm mailing list
>> linux-lvm@redhat.com
>> https://www.redhat.com/mailman/listinfo/linux-lvm
>> read the LVM HOW-TO at http://tldp.org/HOWTO/LVM-HOWTO/
>
>
>
> _______________________________________________
> linux-lvm mailing list
> linux-lvm@redhat.com
> https://www.redhat.com/mailman/listinfo/linux-lvm
> read the LVM HOW-TO at http://tldp.org/HOWTO/LVM-HOWTO/

Re: [linux-lvm] How does LVM-Cache work?

[thornber]

On Wed, Feb 27, 2013 at 03:30:47PM -0300, Leo Antonio wrote:
> Thanks for answers.
> 
> 
> I take a look in dm-cache-target, I really dont knew this project.
> 
> Joe, what is the best way to continue this discussion?
> 
> Do you have some documentation besides what is into kernel source, like
> Papers, Architecture Projects etc?
> What is the best way to start to understand this?

The best place to start is by reading
Documentation/device-mapper/cache.txt and trying it out a bit.  Then
read drivers/md/dm-cache-policy.h which defines the interface that
policies use.  Finally look at dm-cache-policy-mq.c (multiqueue
policy).  This should give you a good idea of the responsibilities of
a policy.

An alternative project would be creating a framework for assessing
cache policies.  Not just providing a range of different cache loads
(though that is a big important part), but also providing feedback to
the policy developer as to how their policy could be improved.  I'd like
to see a tool that calculates the optimimum series of
promotions/demotions for a given io load and so comes up with a
theoretical 'best' performance.

- Joe