[pnfs] Linux pNFS road map

[Dean Hildebrand]

J. Bruce Fields wrote:
> On Wed, Dec 12, 2007 at 02:54:13PM -0500, peter honeyman wrote:
>   
>> marc,
>>
>> the client may elect to send I/O requests to the MDS for reasons other  
>> than failure, e.g., small requests.
>>     
>
> And you'd like to be able to plug in your non-pNFS server and expect at
> least some minimal functionality.  I assume that's the main reason the
> spec requires this.
>   
The other way is required as well, you would like to use your 
NFSv4.1/NFSv4 client and expect some level of functionality.  This is 
required by the minor-versioning aspect, non?.

So spNFS is only partially in the kernel.  In order to be acceptable to 
linux maintainers, how much code must reside in the kernel?  Must it be 
ported to the kernel before it is accepted?

I guess once spNFS is released, and we can see the todo list, maybe 
we'll have a better understanding of how much needs to be done before it 
would be acceptable to linux maintainers.  Once we have that 
information, we will need to look at the tradeoff between finishing 
spNFS and working towards linux acceptance via a file-based 
implementation with GFS2.

Dean