Best design docs for XFS

Best design docs for XFS

[John Quigley]

Hey Folks:

What are the best and most current technical design documents pertaining to (Linux port of) XFS, as a companion to reading the actual code?  Based upon what I've read on the community site:

http://xfs.org/index.php/XFS_Papers_and_Documentation

the most thorough technical documentation of the system lives at:

http://oss.sgi.com/projects/xfs/design_docs/xfsdocs93_pdf/

My only concern with the above is the age - are these sufficiently up-to-date from an architecture perspective?  Thanks very much.

- John Quigley

home: www.jquigley.com

_______________________________________________
xfs mailing list
xfs@oss.sgi.com
http://oss.sgi.com/mailman/listinfo/xfs

Re: Best design docs for XFS

[Eric Sandeen]

John Quigley wrote:
> Hey Folks:
> 
> What are the best and most current technical design documents
> pertaining to (Linux port of) XFS, as a companion to reading the
> actual code?  Based upon what I've read on the community site:
> 
> http://xfs.org/index.php/XFS_Papers_and_Documentation
> 
> the most thorough technical documentation of the system lives at:
> 
> http://oss.sgi.com/projects/xfs/design_docs/xfsdocs93_pdf/
> 
> My only concern with the above is the age - are these sufficiently
> up-to-date from an architecture perspective?  Thanks very much.

It's pretty good still, the fundamental design is not substantially
changed; you're more likely to find missing bits than "wrong" bits I think.

http://oss.sgi.com/projects/xfs/training/index.html

has more info as well.

-Eric

_______________________________________________
xfs mailing list
xfs@oss.sgi.com
http://oss.sgi.com/mailman/listinfo/xfs

Re: Best design docs for XFS

[Christoph Hellwig]

On Mon, Aug 10, 2009 at 05:56:23PM -0500, Eric Sandeen wrote:
> It's pretty good still, the fundamental design is not substantially
> changed; you're more likely to find missing bits than "wrong" bits I think.
> 
> http://oss.sgi.com/projects/xfs/training/index.html
> 
> has more info as well.

The fundamental design is mostly unchanged.  What you might now want to
bother reading of the original documents are:

 dir_design.pdf - the dir code has been completely replaced
 threads.pdf - this is totally IRIX specific
 scaffold.pdf - again mostly IRIX specific
 message.pdf - no idea what this is at all, I never saw traces of this
 in XFS

For all others there's quite a bit of IRIXism in there.

Is there anything specific you're looking for?


> 
> -Eric
> 
> _______________________________________________
> xfs mailing list
> xfs@oss.sgi.com
> http://oss.sgi.com/mailman/listinfo/xfs
---end quoted text---

_______________________________________________
xfs mailing list
xfs@oss.sgi.com
http://oss.sgi.com/mailman/listinfo/xfs

Re: Best design docs for XFS

[John Quigley]

Christoph Hellwig wrote:
> On Mon, Aug 10, 2009 at 05:56:23PM -0500, Eric Sandeen wrote:
>> It's pretty good still, the fundamental design is not substantially
>> changed; you're more likely to find missing bits than "wrong" bits I think.
>>
>> http://oss.sgi.com/projects/xfs/training/index.html
>>
>> has more info as well.
> 
> The fundamental design is mostly unchanged.  What you might now want to
> bother reading of the original documents are:
> 
>  dir_design.pdf - the dir code has been completely replaced
>  threads.pdf - this is totally IRIX specific
>  scaffold.pdf - again mostly IRIX specific
>  message.pdf - no idea what this is at all, I never saw traces of this in XFS

Thanks a lot guys, that helps.

> For all others there's quite a bit of IRIXism in there.

Ack.

> Is there anything specific you're looking for?

In my work, I've just begun to use XFS as the interface into a very large, virtual storage system, to great effect.  It's my first direct interaction with the file system, and I've been hugely impressed with its power, speed and versatility.

The only unresolved problem that we're experiencing has to do with file system corruption in a failover scenario, and my immediate need in understanding some of the internals is to wrap my hands around this problem (I may email this list with details if I can sufficiently explain myself).

A more long-term need we have is for clustering support.  I've read lightly about CXFS, though it appears to be inactive, or at least not open source.  Has clustering support been discussed with the members here?  If so, I'd like to take part in those conversations if there's sufficient interest to engage in that way.

Thanks very much again.

John Quigley
jquigley.com

_______________________________________________
xfs mailing list
xfs@oss.sgi.com
http://oss.sgi.com/mailman/listinfo/xfs

Re: Best design docs for XFS

[Eric Sandeen]

John Quigley wrote:
> Christoph Hellwig wrote:

...

>> Is there anything specific you're looking for?
> 
> In my work, I've just begun to use XFS as the interface into a very
> large, virtual storage system, to great effect.  It's my first direct
> interaction with the file system, and I've been hugely impressed with
> its power, speed and versatility.
> 
> The only unresolved problem that we're experiencing has to do with
> file system corruption in a failover scenario, and my immediate need
> in understanding some of the internals is to wrap my hands around
> this problem (I may email this list with details if I can
> sufficiently explain myself).

Any failover you're doing would need to be handled outside the
filesystem, of course; STONITH etc?  I don't know that reading xfs
internals will be all that relevant.

But I suppose the details may help if you think it is an xfs-related
problem.

> A more long-term need we have is for clustering support.  I've read
> lightly about CXFS, though it appears to be inactive, or at least not
> open source.  Has clustering support been discussed with the members
> here?  If so, I'd like to take part in those conversations if there's
> sufficient interest to engage in that way.

CXFS is and probably will be the only direct clustering mechanism for
xfs... it is still active, though as you note it is not (and probably
will never be?) open source.

block-layer failover should be as possible on xfs as any other
filesystem, though.

-Eric

> Thanks very much again.
> 
> John Quigley jquigley.com
> 

_______________________________________________
xfs mailing list
xfs@oss.sgi.com
http://oss.sgi.com/mailman/listinfo/xfs