NFS v4 Working Group Minutes

First Day July 29:

Agenda:

DN Introduction, note-well, agenda-bashing. Discussion of note-
taking.
DN 5661bis

1. Note well. Blue sheets.

No comments. (we’re having audio problems with DN) Blue Sheet sign in is automatic for virtual meetings.

2. 5661bis

See slides.

Nico did a I18N review. I18N for file system is different than app level. We need I18N expertise to move this forward.

Security should be manageable effort. To be discussed in next section.

5661bis-00 expired. -01 to be posted next week. Focus of work is on document split up. Interdocument references will be major part of the split-up effort for next couple of months.

Getting to last call, splitting up document. Six documents need editors.

Sorin: A draft RFC describing the split would be useful…

dn: -01 will describe split-up.

3. Security

DN presents. See slides.

DN describing base issues with security (which sound eerily familiar to original issues with NFS years ago).

SHOULD and MUST motivations in ACL and mode bits section. Outdated POSIX ACL model references are pointless today. Never adopted.

-01 doc should be available in a month.

There are SHOULDs regarding handling of the attributes, not motivated.

4. RPC-with-TLS

CL presents. See slides.

Clarity rename to RPC with TLS.

Need a smoke test for RPC-with-TLS. Discuss details offline (Dickson and Lever). Discussions with Redhat on QA.

Linux implementation experience discussion - does not impact protocol.

Second Day July 30

**1a Note Well

Note Well reiterated. Blue Sheet sign in is automatic for virtual meetings.

5. QUIC

See slides. CL presents.

QUIC provides TLS support - but we are adding that. CL says he is looking for use cases for QUIC.

Congestion controol is more robust than TCP - maybe not needed for NFS, because it runs on data center networks.

QUIC user space implementations are problematic for kernel space transport service consumers like NFS.

MS QUIC - Nick Banks asking for clarification on loss of performance. Discussion on internet throughput vs. high speed data center network - CL.

Richard Scheffenegger says current HW can not fully offload crypto for QUIC (yet).

https://microsoft.github.io/msquic/ has our perf data

James Fields asks “Isn’t there any per-stream overhead?” Nick Banks replied “Some. But it’s low enough that some protocols use a new stream for each audio/video frame in streaming protocols.”

Take this discussion on benefits of QUIC work to mail list. - Zaheduzzaman Sarker NFSV4 AD.

6. RoR v2

See slides.

CL described challenges with credit-based flow control section of current document. It is not implementable.

Jana Iyengar has not responded to requests for expert review. Zaheduuzzaman offered to help look for other experts.

7. Signing

See slides.

Merkle Trees for digest.

8. Future Work

Move to mail list this discussion.

Expand allBack to topGo to bottom