Re: [nfsv4] Slides from the pNFS BoF we delivered at LISA 11 - with attachment this time

[Boaz Harrosh <bharrosh@panasas.com>]

On 12/09/2011 04:53 AM, Rick Macklem wrote:
> Thanks for the slides, Sorin.
> 
> Just fyi, it was correct to not mention a FreeBSD client, since the
> 4.1 client code hasn't made it into any release yet. It is available
> as a patch against the FreeBSD9.0 kernel, in case anyone is interested:
>    http://people.freebsd.org/~rmacklem/nfsv4.1-client/nfsv41.patch
> 
> I am just starting to work on the logistics of getting this into the
> FreeBSD source tree, so it may be in a FreeBSD9.n release, where n > 0.
> 
> Also, the above does not include any pNFS support. I am about ready to
> start coding pNFS client stuff, which brings me to...
> - It will be difficult for me to make much progress without any
>   pNFS servers to test against. (The only NFSv4.1 server I have to
>   test against is Fedora15 and I don't think it knows how to do pNFS?
>   If there is a way to make Fedora15/16 (or any other freebie Linux
>   distro) do pNFS, please let me know.
> 
> If others have suggestions on how I might gain access to pNFS servers
> for testing, I would appreciate hearing. (However, I am not funded to
> do any of this, so the suggestions can't include my spending $$$. I can
> make it to a Bakeathon when it happens to be close to where I am, so I
> can show up "on the cheap", like I did last June in Ann Arbor.)
> 

One way to go is the Linux pnfs project at linux-nfs.org start here:
http://wiki.linux-nfs.org/wiki/index.php/PNFS_Development_Git_tree

It is an NFSD linux Kernel server with pNFS extensions destined to
be part of Linux at some future release.

You will need to compile your own Kernel, or alternatively this Kernel
is available in rpm form  from Steve Dickson for some Fedoras:
http://steved.fedorapeople.org/repos/pnfs/

The easiest is to first use the what is called pnfs-exp which is an MDS
plus a single DS exporting any local file system.

Then you can also export a GFS2 read-only and an EXOFS which is a full
fixture pnfs-objects based cluster filesystem.

The Luster patch suggested to you is also based on above Benny's tree.

I have on my todo list a full osd (pnfs-objects) stack port to FreeBSD
But that's in the far future. 
Meanwhile I'd suggest the pnfs-client-core to layout-driver API be closely
crafted to match that of the Linux LD API so it could be kept the same
code base. For example the pnfs-objects LD is mostly pure C code relying
on the ORE (Objects Raid Engine). The ORE is multy licensed library that
has ports to Many platforms including Linux-Kernel 
Linux/FreeBSD/windows User-Mode and more of which only the Linux variants
actually run because of the need of the iscsi stack to support BiDi and
large CDB commands, Only supported by Linux.

Cheers
Boaz

> rick
> 
> ----- Original Message -----
>> On Thu, 08 Dec 2011 19:40:19 -0500, Sorin Faibish <sfaibish@emc.com>
>> wrote:
>>
>>> Attached please find the slides that we presented yesterday at
>>> USENIX
>>> LISA 11
>>> conference in Boston. We would like to get more detailed input on
>>> other
>>> OS
>>> that will support pNFS server/client (this is for Rick as well) that
>>> were
>>> not mentioned in this slide set.
>>>
>>> /Sorin
>>>
>>
>>
>>
>> --
>> Best Regards
>> Sorin Faibish
>> Corporate Distinguished Engineer
>> Unified Storage Division
>>
>> EMC²
>> where information lives
>>
>> Phone: 508-435-1000 x 48545
>> Cellphone: 617-510-0422
>> Email : sfaibish@emc.com
> _______________________________________________
> nfsv4 mailing list
> nfsv4@ietf.org
> https://www.ietf.org/mailman/listinfo/nfsv4