[LSF/MM TOPIC] VFS hot tracking

[LSF/MM TOPIC] VFS hot tracking

[Zhi Yong Wu]

HI,

   If some guy is interested in this topic, i would like to talk about
VFS hot tracking.

   The feature will add the support for tracking data temperature
information in VFS layer. Essentially, this means maintaining some key
hot data statistics (like number of reads/writes, last read/write
time, frequency of reads/writes), then distilling those numbers down
to a single "temperature" value that reflects what data is "hot", then
specific FS such as btrfs will take use of the temperature infomation
to move hot data to SSDs, or do the defragmetation.

   In this talk, i would discuss its current implementation and how
its key points are resolved such as memory consumption, how its hot
data is updated periodically, etc.

   If you have any suggestions, pleas let me know, thanks.


-- 
Regards,

Zhi Yong Wu

Re: [LSF/MM TOPIC] VFS hot tracking

[mingming cao]

On 02/07/2013 08:24 AM, Zhi Yong Wu wrote:
> HI,
>
>     If some guy is interested in this topic, i would like to talk about
> VFS hot tracking.
>
>     The feature will add the support for tracking data temperature
> information in VFS layer. Essentially, this means maintaining some key
> hot data statistics (like number of reads/writes, last read/write
> time, frequency of reads/writes), then distilling those numbers down
> to a single "temperature" value that reflects what data is "hot", then
> specific FS such as btrfs will take use of the temperature infomation
> to move hot data to SSDs, or do the defragmetation.
>
>     In this talk, i would discuss its current implementation and how
> its key points are resolved such as memory consumption, how its hot
> data is updated periodically, etc.
>
>     If you have any suggestions, pleas let me know, thanks.
>
>

That sounds a good fit for the LSF discussion. I am happy to hear the 
update of this feature and where it will go from here. Especially 
discussions of APIs that expose the data hotness and how that 
information could be used by filesystems and storages.

Mingming

Re: [LSF/MM TOPIC] VFS hot tracking

[james northrup]

I'm an observer but this topic has been around for a while.

Zhi can you present a slide or three to suggest the tree operations
associated with the heat data management and its persistence
lifecycle?


>> HI,
>>
>>     If some guy is interested in this topic, i would like to talk about
>> VFS hot tracking.
>>
>>     In this talk, i would discuss its current implementation and how
>> its key points are resolved such as memory consumption, how its hot
>> data is updated periodically, etc.
>>
> That sounds a good fit for the LSF discussion. I am happy to hear the update of this feature and where it will go from here. Especially discussions of APIs that expose the data hotness and how that information could be used by filesystems and storages.

Re: [LSF/MM TOPIC] VFS hot tracking

[Zhi Yong Wu]

On Fri, Feb 8, 2013 at 7:48 AM, james northrup <northrup.james@gmail.com> wrote:
> I'm an observer but this topic has been around for a while.
>
> Zhi can you present a slide or three to suggest the tree operations
> associated with the heat data management and its persistence
> lifecycle?
Sure, if this topic is accepted.:)

>
>
>>> HI,
>>>
>>>     If some guy is interested in this topic, i would like to talk about
>>> VFS hot tracking.
>>>
>>>     In this talk, i would discuss its current implementation and how
>>> its key points are resolved such as memory consumption, how its hot
>>> data is updated periodically, etc.
>>>
>> That sounds a good fit for the LSF discussion. I am happy to hear the update of this feature and where it will go from here. Especially discussions of APIs that expose the data hotness and how that information could be used by filesystems and storages.



-- 
Regards,

Zhi Yong Wu

Re: [LSF/MM TOPIC] VFS hot tracking

[Zheng Liu]

On Fri, Feb 08, 2013 at 08:55:02AM +0800, Zhi Yong Wu wrote:
> On Fri, Feb 8, 2013 at 7:48 AM, james northrup <northrup.james@gmail.com> wrote:
> > I'm an observer but this topic has been around for a while.
> >
> > Zhi can you present a slide or three to suggest the tree operations
> > associated with the heat data management and its persistence
> > lifecycle?
> Sure, if this topic is accepted.:)

Hi Zhi Yong,

Maybe a slide with some measurements of performance is better. ;-)

Regards,
                                                - Zheng