Re: The (sorry?) state of pNFS documentation?! Abandonware?

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



On 3/26/25 10:33 AM, Martin Wege wrote:
> On Tue, Feb 11, 2025 at 3:00 PM Chuck Lever <chuck.lever@xxxxxxxxxx> wrote:
>>
>> On 2/11/25 7:54 AM, Martin Wege wrote:
>>> Is there any up to date documentation for pNFS server support in the
>>> Linux 6.6 kernel?
>>
>> There isn't up-to-date documentation for NFSD's pNFS support. There are
>> various efforts going on to improve it, but as we are swamped with other
>> more pressing issues, there hasn't been good progress.
>>
>> pNFS block is supported, but it's not straightforward to set up.
>>
>> pNFS flexfiles is supported, but the implementation supports only the
>> case where the DS and MDS are the same server.
>>
>> NFSD does not implement the other layout types.
> 
> More questions:
> 1. Clarification, please:
> Which layout types are and are not supported:
> LAYOUT4_NFSV4_1_FILES
> LAYOUT4_OSD2_OBJECTS
> LAYOUT4_BLOCK_VOLUME

As I stated above, pNFS block is implemented. "Does not implement the
other layout types" means the NFSV4_1_FILES and OSD2_OBJECTS layout
types are not implemented.

The I/O protocols that pNFS block can use include SCSI, iSCSI, and NVMe.


> 2. Is Flexfiles also part of enum layouttype4, or something different?

Flexfiles is a layout type. It is described in RFC 8435

  https://datatracker.ietf.org/doc/html/rfc8435

NFSD's implementation is simplified and not for production use. It is
useful for testing client implementations of the flexfile layout type
and not much more at this time.


> 3. dCache supports pNFS MetaDataServer (MDS), and NFSv3 Data Servers
> (DS). Where is the spec for this?

This sounds like the flexfiles layout type to me. You will have to
confirm that with Tigran and his team.


> And why, WHY NFSv3 DS? Why not NFSv4.1 DS=

Supporting NFSv4.1 data servers is what the RFC 5661 NFSV4_1_FILES
layout type already does.

For more on the motivation behind flexfiles, consult RFC 8435. But
generally the reasons are:

- NFSv3 READ and WRITE continue to be lower latency than NFSv4

- There's still a lot of data in the world that lives on NFSv3-only
servers. Serving it in place is cheaper than migrating it.


-- 
Chuck Lever




[Index of Archives]     [Linux Filesystem Development]     [Linux USB Development]     [Linux Media Development]     [Video for Linux]     [Linux NILFS]     [Linux Audio Users]     [Yosemite Info]     [Linux SCSI]

  Powered by Linux