Re: [nfsv4] Proposed charter changes (with updates)

[Tom Talpey <[e-mail]>]

On 6/5/2020 9:03 AM, David Noveck wrote:
> 
> 
> On Fri, Jun 5, 2020 at 8:15 AM Tom Talpey <[e-mail] 
> <mailto:[e-mail]>> wrote:
> 
>     On 6/5/2020 3:25 AM, Lars Eggert wrote:
>      > Hi,
>      >
>      > On 2020-6-5, at 10:23, Magnus Westerlund
>     <magnus.westerlund=[e-mail]
>     <mailto:[e-mail]>> wrote:
>      >> For that reason I would recommend that one work through the
>     technical proposal
>      >> to better understand what desired propoerties of RDMA over QUIC
>     has, as well as
>      >> what is required to realize those from QUIC. Then one can likely
>     make a good
>      >> decision if that work should be done in QUIC WG or in this or
>     another WG.
>      >
>      > +1
>      >
>      > An initial individual draft would be very useful in triaging
>     where the best place for this work would be.
> 
>     So, to be clear I was describing a process for a potential
>     NFSv4-over-QUIC binding, not an RDMA-over-QUIC one. 
> 
> 
> That would probably take the form of an extension to RPC, updating RFC 
> 5531, with no change to QUIC itself anticipated.  Unless I'm missing 
> something, that could be done even under the current charter, although 
> I'm cc-ing Lars to make sure he is aware of the possibility of us doing 
> this.

The NFSv4 WG has already published an Upper Layer Binding ("ULB") for
various dialects of NFS atop RPC-over-RDMA, so yes, I agree.

However, with a new transport, in my opinion, there is always the
possibility of finding something "missing", for example, the current
Push Mode extension for RDMA. It's this latter aspect in which I
strongly believe the upper layer, e.g. NFSv4, can add value.

>     Although,
>     I have been thinking about both.
> 
> Apparently, Magnus read your mind.   I'm not sure how I feel about 
> having AD's with that capability :-)
> 
>     I'd like to get the RDMA Push Mode extension draft process
>     through the knothole before considering the next one. :-)
> 
> 
> I hope "through the knothole" means waiting for IESG approval.   Waiting 
> for publication as an RFC has lately become a process of unbounded duration.

Achieving IESG approval, yes. That's why I added the word "process".
Once approved, we can start the actual work.

> I think you are the right person to do the individual draft for 
> RDMA-over-QUIC and hope that you would be interested in doing it, when 
> the time is ripe.
> 
> With regard to RPC-over-QUIC, I'd be glad if you are interested in doing 
> this, but there are a number of people that can participate so it would 
> not take all that much of your time.  It is interesting to the working 
> group primarily because of its security implications.

I'm flattered, and thanks. I'm not unique in having thought about it
though. Earlier, it was clear that standardizing a base QUIC protocol
was needed before considering this. As QUIC moves forward, I'd be hopeful
that it makes sense to explore RDMA.

Tom.