IETF-117 NFSv4 Agenda
Welcome (back) to San Francisco

•	Chairs Welcome / Agenda Bash (chairs)

https://datatracker.ietf.org/meeting/117/materials/slides-117-nfsv4-chair-slides/
https://datatracker.ietf.org/meeting/117/materials/slides-117-nfsv4-noveck-rfc5661bis/
https://datatracker.ietf.org/meeting/117/materials/slides-117-nfsv4-noveck-secure-rdma/

•	WG status (10 min - chairs / all)

	◦	interim meeting report
	◦	document status update
	◦	list agreement / status

beepy did doc report. We live adopted the Tom Talpey documents into the WG. Tom will submit drafts for WG moving forward.

5661-bis - Hellwig asked if we really need the -bis document, what problems are being solved.

Are vendors seeing security asks for NFS. Haynes says they are asked about security but not about missing features in -bis document. Haynes - can do security with TLS or Kerberos today.

NB: Sorin had to restart meet echo three times. Meetecho continues to be flaky for remote participants as it was during our interim meeting.

•	security approach (15 min - all)


•	internationalization status (10 min - all)

Lars Eggert - following up on Klensin comments. Chris and Zahed and Dave Novice and Klensin kick off meeting next week. 

Hellwig thinks current internationalization document is not ready for review. In practice, implementations don’t do internationalization per IETF. Existing published versions of NFS client and server do not do internationalization. It is done by underlying file systems/layers of the implementation outside the protocol. (as was mentioned in interim meeting - similar to FTP).

Hellwig suggests not even trying to clarify implementations.

•	RDMA Security (10 min - Noveck)

Lars Eggert gave a rundown of QUIC status:

	◦ Scale to many streams
	◦ Good security
	◦ Head of line blocking per stream
	◦ Think about library to use implementation-wise 
	◦ Multipath not ready

Lars responded to Dave’s comments with regards to QUIC and security. Hellwig - skin the cat from from a different direction. Not network, but storage - people use RDMA to properly align the data payload and HW offload. How do we get a protocol … (look up transcript). Really need HW implementation support - that's the RDMA advantage.

Tom said we looked at layering iWarp on QUIC - but not ready at time. Would unfortunately pick up a lot of TCPisms, and it needs a lot of iWarp layer work. Tom has proposal to layer encryption without protocol change to only encrypt data payload. Let’s backup and start with requirements.

Lars is back. While we have hardware offload for TLS, bit not HW offload for QUIC. (look at transcript))

Hellwig says we need to talk to the HW vendors. EBPF context discussions. Pager splitting & alignment is on many wishlists, but it's hard. Network protocol state is hard, might be smart to use lower state.

Tom - iWarp over TCP over TLS - NIC vendors not very enamoured of it. Difficult to provide. Second, SMB use of QUIC is a lousy example. Not at all performant.Tabled this before, likely need QUIC tweak. Let's think about requirements.

Lars said Talpey points are good. Christoph point on HW offload is true today, but QUIC offload will happen eventually.

Q Misell says getting QUIC in HW offload would be great. But don’t disregard QUIC because no HW offload.

Noveck says we can slip into existing RFCs - see slides.

Dave Noveck says there is path to software implementation. Worth doing an experiment.

Lars said 20 or so QUIC implementations including BSD licensed from Microsoft (kernel and client compilable from same source?), something about Linux. Microsoft use of QUIC on SMB is wide use.

QUIC for NFS would have its uses regardless, at least for WAN. Waiting for HW offload probably not necessary.

Talpey says we’re underestimating the work to exploit QUIC. See transcript.

•	rfc5661bis (* - Noveck)

Updated slides from interim meeting. See slide on motivations.

Really have to go through transcript.

Hellwig says changing pNFS terminology not helpful at this time. Haynes thinks 8434 defines this terminology. Verify https://datatracker.ietf.org/doc/rfc8434/

Chris Inacio - David Black said we should change RECOMMENDED to SHOULD or REQUIRED? On attributes.

No argument to cleaning up wording to be IETF normal reflecting what we intended. Not a problem now, Lars, prevent future misunderstanding.

Q Misell - for future audiences. Clean up.

* Chris notes: security plan doesn't work as well when we use block protocols via pNFS - they are outside IETF security domain, and have their own security. Chris wants to do a security analysis on pNFS.

•	way forward / planning (10 min - all)

Get interim meetings on people calendars. 

Richard S brought up connect issue - should this be a working agreement. Known best ways document to handle? 

Lars has left the room.

•	NVME draft - Agenda addition by Hellwig

Initial implementer feedback on NVME feedback. Implementations want to use the UUID identifier. One page draft + boiler plate. Hellwig to submit draft.

Richard S said has to go to meeting materials (check transcript).