Lustre devel - Jan 2011 - How can the name_to_handle() syscall benefit Lustre

Hi All,

How can the name_to_handle() syscall benefit Lustre ?

Any insights will be helpful.

Thanks,
Vilobh
*Graduate Research Associate
Department of Computer Science
The Ohio State University Columbus Ohio*
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
http://lists.lustre.org/pipermail/lustre-devel/attachments/20110129/8ce8b36d/attachment.html

This is something that I think would be very useful for Lustre in conjunction
with the Lustre MPI ADIO driver. The MPI rank-0 node could do the path traversal
and generate a file handle via name_to_handle() and then MPI broadcast the file
handle to all of the other MPI processes, which can use it to open the shared
file. This could save many thousands of MDS RPCs to do path traversal for each
client in a large MPI job.

Cheers, Andreas

On 2011-01-29, at 12:16, Vilobh Meshram <vilobh.meshram at gmail.com>
wrote:
> Hi All,
> 
> How can the name_to_handle() syscall benefit Lustre ?
> 
> Any insights will be helpful.
> 
> Thanks,
> Vilobh
> Graduate Research Associate
> Department of Computer Science
> The Ohio State University Columbus Ohio
> _______________________________________________
> Lustre-devel mailing list
> Lustre-devel at lists.lustre.org
> http://lists.lustre.org/mailman/listinfo/lustre-devel
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
http://lists.lustre.org/pipermail/lustre-devel/attachments/20110130/2d7eaa4a/attachment.html

> This is something that I think would be very useful for Lustre in
conjunction with the Lustre MPI ADIO driver. The MPI rank-0 node
> could do the path traversal and generate a file handle via name_to_handle()
and then MPI broadcast the file handle to all of the
> other MPI processes, which can use it to open the shared file. This could
save many thousands of MDS RPCs to do path traversal for
> each client in a large MPI job.?
I thought the name_to_handle syscall idea was shot down? 
 
> Cheers, Andreas
> 
> On 2011-01-29, at 12:16, Vilobh Meshram <vilobh.meshram at gmail.com>
wrote:
> 
>       Hi All,
> How can the name_to_handle() syscall benefit Lustre ?
> 
> Any insights will be helpful.
> 
> Thanks,
> Vilobh
> Graduate Research Associate
> Department of Computer Science
> The Ohio State University Columbus Ohio
> 
>       _______________________________________________
>       Lustre-devel mailing list
>       Lustre-devel at lists.lustre.org
>       http://lists.lustre.org/mailman/listinfo/lustre-devel
> 
> 
>

On 2011-01-31, at 09:52, James Simmons wrote:
> Andreas Dilger wrote:
>> This is something that I think would be very useful for Lustre in
conjunction with the Lustre MPI ADIO driver. The MPI rank-0 node
>> could do the path traversal and generate a file handle via
name_to_handle() and then MPI broadcast the file handle to all of the
>> other MPI processes, which can use it to open the shared file. This
could save many thousands of MDS RPCs to do path traversal for
>> each client in a large MPI job. 
> 
> I thought the name_to_handle syscall idea was shot down? 
I don''t think it was shot down, as much as it hasn''t yet been
accepted.  I''ve tried to support it in a form that would be useful to
Lustre/MPI, and while it isn''t quite where I''d like it to be,
I think the flaws in the implementation are not insurmountable.

The current proposal is lacking a well-defined permanent filesystem identifier
in each handle (i.e. a UUID as part of the public fh structure), but I think
Lustre can still include the UUID as part of the private fh and have the
userspace code map that back to a specific Lustre mountpoint, since it would be
implemented as part of the Lustre ADIO layer.
>> On 2011-01-29, at 12:16, Vilobh Meshram <vilobh.meshram at
gmail.com> wrote:
>> How can the name_to_handle() syscall benefit Lustre ?
>> 
>> Any insights will be helpful.
>> 
>> Thanks,
>> Vilobh
>> Graduate Research Associate
>> Department of Computer Science
>> The Ohio State University Columbus Ohio
>> 
>>      _______________________________________________
>>      Lustre-devel mailing list
>>      Lustre-devel at lists.lustre.org
>>      http://lists.lustre.org/mailman/listinfo/lustre-devel
>> 
>> 
>> 

Cheers, Andreas
--
Andreas Dilger 
Principal Engineer
Whamcloud, Inc.