dmapi support for reiser4

dmapi support for reiser4

[Bjoern JACKE]

Hi,

DMAPI support is a feature which is essential for usage in many 
important environments. Are there any plans to add dmapi support to 
reiser4 ?

Bjoern

Re: dmapi support for reiser4

[Vladimir Saveliev]

Hello

On Tue, 2005-06-28 at 13:34, Bjoern JACKE wrote:
> Hi,
> 
> DMAPI support is a feature which is essential for usage in many 
> important environments. Are there any plans to add dmapi support to 
> reiser4 ?
> 
there is no currently

> Bjoern
> 

Re: dmapi support for reiser4

[Bjoern JACKE]

On 2005-06-28 at 15:31 +0400 Vladimir Saveliev sent off:
>> DMAPI support is a feature which is essential for usage in many 
>> important environments. Are there any plans to add dmapi support to 
>> reiser4 ?
>> 
>there is no currently

would be very nice to have that, one day reiser4 has survived all the 
troubleful discussion about kernel integration it is in right now. 
DMAPI is really a feature which is a must-have for big setups. One of 
the reasons is file allocation info which is retrievable which makes 
sparse file handling efficient. Of course there are much more useful 
posibilities DMAPI gives to get over POSIX limitation. XFS, JFS2, 
GPFS, Veritas filesystem for Solaris, they all have DMAPI support 
and reiser4 will not be a real competitor without this. I hope it 
will come one day ...

Bjoern

Re: dmapi support for reiser4

[Vladimir Saveliev]

Hello

On Wed, 2005-06-29 at 15:57, Bjoern JACKE wrote:
> On 2005-06-28 at 15:31 +0400 Vladimir Saveliev sent off:
> >> DMAPI support is a feature which is essential for usage in many 
> >> important environments. Are there any plans to add dmapi support to 
> >> reiser4 ?
> >> 
> >there is no currently
> 
> would be very nice to have that, one day reiser4 has survived all the 
> troubleful discussion about kernel integration it is in right now. 
> DMAPI is really a feature which is a must-have for big setups. One of 
> the reasons is file allocation info which is retrievable which makes 
> sparse file handling efficient. Of course there are much more useful 
> posibilities DMAPI gives to get over POSIX limitation. XFS, JFS2, 
> GPFS, Veritas filesystem for Solaris, they all have DMAPI support 
> and reiser4 will not be a real competitor without this. I hope it 
> will come one day ...
> 

I would guess that it is rather operating system feature than a
filesystem one. Can you please point some source of information about
DMAPI?

> Bjoern
> 

Re: dmapi support for reiser4

[Bjoern JACKE]

Hi,
On 2005-06-29 at 20:15 +0400 Vladimir Saveliev sent off:
>I would guess that it is rather operating system feature than a
>filesystem one. Can you please point some source of information about
>DMAPI?

the main part is to be done by the filesystem, the OS cannot do 
everything by itself, there a plenty of things which just the 
filesystem can do by itself. Here are some links:

http://www.opengroup.org/onlinepubs/9657099/
http://www.computer.org/conferences/mss95/lawthers/lawthers.htm
http://www.csu.edu.au/special/auugwww96/proceedings/alex/alex.html

You can also take a look at the XFS and JFS source code parts of dmapi to get
an impression of what is to be done for it at fs layer.

Bjoern

Re: dmapi support for reiser4

[Chet Hosey]

Bjoern JACKE wrote:

> Hi,
> On 2005-06-29 at 20:15 +0400 Vladimir Saveliev sent off:
>
>> I would guess that it is rather operating system feature than a
>> filesystem one. Can you please point some source of information about
>> DMAPI?
>
>
> the main part is to be done by the filesystem, the OS cannot do
> everything by itself, there a plenty of things which just the
> filesystem can do by itself. Here are some links:
>
> http://www.opengroup.org/onlinepubs/9657099/
> http://www.computer.org/conferences/mss95/lawthers/lawthers.htm
> http://www.csu.edu.au/special/auugwww96/proceedings/alex/alex.html
>
> You can also take a look at the XFS and JFS source code parts of dmapi
> to get
> an impression of what is to be done for it at fs layer.
>
> Bjoern

Is there a standard VFS interface that these filesystems present, or is
it done via filesystem-specific extensions? If the latter, perhaps you
should consider submitting a kernel patch that adds the required
interface to the VFS layer, and note in your submission that XFS and JFS
aren't compliant.

How is DMAPI different from FAM with an extra API function for creating
explicit holes?

You said that "reiser4 will not be a real competitor without this". A
competitor in what market? Can you name an application that depends on
DMAPI and is so important that without that particular application
Reiser4 is not appropriate for storing files?

I note that the second link you provided is ten years old, nearly to the
day. The third is nine years old.I'm not suggesting that you're
overstating the importance of DMAPI, but why do you think that something
so important might have been overlooked for so long? Novell just
switched to Reiserfs as their default SUSE filesystem. If Reiserfs isn't
a "real competitor" without DMAPI, why would they do something so unwise?

Just curious, as I'd not heard of DMAPI prior to your post, but you
claim it's quite important.

-- chet

Re: dmapi support for reiser4

[Ragnar Kjørstad]

On Wed, Jun 29, 2005 at 06:41:46PM -0400, Chet Hosey wrote:
> >> I would guess that it is rather operating system feature than a
> >> filesystem one. Can you please point some source of information about
> >> DMAPI?
> 
> Is there a standard VFS interface that these filesystems present, or is
> it done via filesystem-specific extensions? If the latter, perhaps you
> should consider submitting a kernel patch that adds the required
> interface to the VFS layer, and note in your submission that XFS and JFS
> aren't compliant.

The OpenXDSM project intended to extend the VFS layer with a few extra
functions and to implement a generic XDSM/DMAPI layer on top of that.

The project never really got anywhere and the sourceforge webpage has
been removed, but you can still find some pointers to the VFS
discussions if you google for OpenXDSM.

In short, the filesystems needed to be extended with functionality in
two primary areas:
- persistent storage of metadata
- beeing able to punch holes in existing files.

The first item has since been solved by extended attributes, and I
believe there were some ideas about how to change truncate to also be
able to punch holes. There were probably a few more things needed to be
done, but I don't remember. Google probably knows.

> You said that "reiser4 will not be a real competitor without this". A
> competitor in what market? Can you name an application that depends on
> DMAPI and is so important that without that particular application
> Reiser4 is not appropriate for storing files?

DMAPI is used for Hirarchial Storage Management, migrating unused files
to slower medium such as tape. It can also be used by backupsystems.

I should add though, that not everybody that does HSM are big fans of
DMAPI. It's quite possible that it would be just as productive to add a
_different_ API for HSM to linux as adding DMAPI.

> Just curious, as I'd not heard of DMAPI prior to your post, but you
> claim it's quite important.

It is important in a relatively narrow (high end) market, not really for
desktops and low end servers.


-- 
Ragnar Kjørstad

Re: dmapi support for reiser4

[Chet Hosey]

Ragnar Kjørstad wrote:

>On Wed, Jun 29, 2005 at 06:41:46PM -0400, Chet Hosey wrote:
>  
>
>>Is there a standard VFS interface that these filesystems present, or is
>>it done via filesystem-specific extensions? If the latter, perhaps you
>>should consider submitting a kernel patch that adds the required
>>interface to the VFS layer, and note in your submission that XFS and JFS
>>aren't compliant.
>>    
>>
>
>The OpenXDSM project intended to extend the VFS layer with a few extra
>functions and to implement a generic XDSM/DMAPI layer on top of that.
>
>The project never really got anywhere and the sourceforge webpage has
>been removed, but you can still find some pointers to the VFS
>discussions if you google for OpenXDSM.
>
>  
>
In short, no, there's no standard VFS implementation. Gotcha.

>In short, the filesystems needed to be extended with functionality in
>two primary areas:
>- persistent storage of metadata
>- beeing able to punch holes in existing files.
>
>The first item has since been solved by extended attributes, and I
>believe there were some ideas about how to change truncate to also be
>able to punch holes. There were probably a few more things needed to be
>done, but I don't remember. Google probably knows.
>  
>
So essentially you're looking for an interface by which you can produce
holes in files? That sounds like a minor feature request, certainly one
without which a filesystem might still be a "real competitor" in a
general market.

>>You said that "reiser4 will not be a real competitor without this". A
>>competitor in what market? Can you name an application that depends on
>>DMAPI and is so important that without that particular application
>>Reiser4 is not appropriate for storing files?
>>    
>>
>
>DMAPI is used for Hirarchial Storage Management, migrating unused files
>to slower medium such as tape. It can also be used by backupsystems.
>
>I should add though, that not everybody that does HSM are big fans of
>DMAPI. It's quite possible that it would be just as productive to add a
>_different_ API for HSM to linux as adding DMAPI.
>
>  
>
Let me clarify -- you made claims that reiser4 isn't a "real competitor"
without DMAPI. Specifically, DMAPI. I asked whether you can name a
DMAPI-dependent application which *must* work on a given filesystem in
order for it to be a "real competitor". You haven't thus far. Now you're
saying, well, not necessarily DMAPI, but *something*.You're being vague.

To publicly claim that Reiser4 isn't a "real competitor" without
your-favorite-buzzword without providing some *actual evidence* as to
why this is the case seems a little bit rude. Can you support your
accusations?

>>Just curious, as I'd not heard of DMAPI prior to your post, but you
>>claim it's quite important.
>>    
>>
>
>It is important in a relatively narrow (high end) market, not really for
>desktops and low end servers.
>
>  
>
Again, I'd consider it polite to either clarify or adjust your claims
that Reiser4 isn't a "real competitor" (note that you didn't initially
name *any* market, you just made a broad statement that Reiser4 can't
compete).

-- chet

Re: dmapi support for reiser4

[Hans Reiser]

Bjoern JACKE wrote:

> On 2005-06-28 at 15:31 +0400 Vladimir Saveliev sent off:
>
>>> DMAPI support is a feature which is essential for usage in many
>>> important environments. Are there any plans to add dmapi support to
>>> reiser4 ?
>>>
>> there is no currently
>
>
> would be very nice to have that, one day reiser4 has survived all the
> troubleful discussion about kernel integration it is in right now.
> DMAPI is really a feature which is a must-have for big setups. One of
> the reasons is file allocation info which is retrievable which makes
> sparse file handling efficient. Of course there are much more useful
> posibilities DMAPI gives to get over POSIX limitation. XFS, JFS2,
> GPFS, Veritas filesystem for Solaris, they all have DMAPI support and
> reiser4 will not be a real competitor without this. I hope it will
> come one day ...
>
> Bjoern
>
>
All it requires is money...... :-/  If you have some knowledge of
someone who might fund it, let me know.