Question about supporting RDMA Extensions for PMEM

Question about supporting RDMA Extensions for PMEM

[Li, Hao]

Hi,

I have noticed that IETF has released a draft of RDMA Extensions for
PMEM [1]. Does libibverbs has a plan to implement these extensions? Are
there some good starting points if we want to participate the development.
Thanks!

[1]: https://tools.ietf.org/id/draft-talpey-rdma-commit-01.html

Regards,
Hao Li

Re: Question about supporting RDMA Extensions for PMEM

[Tom Talpey]

On 10/12/2020 4:13 AM, Li, Hao wrote:
> Hi,
> 
> I have noticed that IETF has released a draft of RDMA Extensions for
> PMEM [1]. Does libibverbs has a plan to implement these extensions? Are
> there some good starting points if we want to participate the development.
> Thanks!
> 
> [1]: https://tools.ietf.org/id/draft-talpey-rdma-commit-01.html

The draft you refer to is an individual draft, not an official IETF
release. However, after a lengthy process, the effort is now an official
work item of the NFSv4 working group, since the original STORM WG was
shut down an alternative process had to be determined.

An updated document is in the works, hopefully in October, and may
become a full RFC as the IETF process advances. My coauthors and I look
forward to this.

In the meantime, the IBTA LWG took up a similar task, and we completed
it several months ago. I am not sure when it might be released as an
official Annex, but I assume that is in the works as well. The IBTA
version of the extensions is semantically similar, but does not include
the "Verify" operation which in the iWARP document. Perhaps this will
be added later.

In theory, the IBTA SWG is in control of specifying any Verbs changes.
The Annex does discuss these, but SWG would still need to ratify. If
Fujitsu is an IBTA member, I'd encourage you to support this. IETF, of
course, has no dependency on IBTA, or Verbs, so participating there is
a separate matter. I encourage that participation, too!

Tom.

Re: Question about supporting RDMA Extensions for PMEM

[Jason Gunthorpe]

On Mon, Oct 12, 2020 at 10:26:32PM -0400, Tom Talpey wrote:

> In theory, the IBTA SWG is in control of specifying any Verbs changes.

SWG and IETF should agree on what the general software presentation
should look like so HW implementations can be compatible.

It looks fairly straightforward so this probably isn't strictly
necessary once the one the wire protocol is decided.

verbs has a life of its own these days outside IBTA/IETF..

Jason

Re: Question about supporting RDMA Extensions for PMEM

[Tom Talpey]

On 10/16/2020 6:37 PM, Jason Gunthorpe wrote:
> On Mon, Oct 12, 2020 at 10:26:32PM -0400, Tom Talpey wrote:
> 
>> In theory, the IBTA SWG is in control of specifying any Verbs changes.
> 
> SWG and IETF should agree on what the general software presentation
> should look like so HW implementations can be compatible.
> 
> It looks fairly straightforward so this probably isn't strictly
> necessary once the one the wire protocol is decided.
> 
> verbs has a life of its own these days outside IBTA/IETF..

For the record, I completely agree with the goal of compatible
interfaces. And, I'm committed to defining it, however I am not
able to participate in IBTA as I am no longer associated with
any IBTA member company. I am active in IETF, which has no such
restriction.

However, it's also important to point out that IETF considers
programming interfaces to be out of scope. The draft-hilland-verbs
document for iWARP was not adopted as a work item, and its
content, while extraordinary useful, is not an IETF product.
As an example, consider TCP and Sockets. IETF owns the former,
and has no input on the latter.

In other words, I think it's on us.

Tom.