IETF 99 - Prague - nfsv4 working group
July 20, 2017, 9:30 - 12:00AM
beepy and Spencer Shepler - co-chairs

-- Note Well, Blue Sheets

RFC 8179 note well

Blue sheets - note your organization. New Note Well RFC - please review.

-- Agenda bashing (beepy)

Additions?

Add "Next Steps" from Haynes and Trond. And Noveck will do the recharter presentation.

-- draft-ietf-nfsv4-migration-issues discussion (includes Adamson's trunking notes) (Noveck) (30 minutes)

Looking to close document that addresses details of migration from 4.0 to 4.1 to 4.2. See slides for details of the migration issues. Trunking issues in particular exist.

v4.1 state migration issues uncovered by Oracle experiments. Migration in RFC5661 - Noveck addressed issues.

Session migration now dealt with in the specification.

Changing of the handling of trunking described in slides... trunking vs. replica distinction clarifications. Dave Noveck would like people to review that the implementations conform to id sharing semantics (verify this with Noveck - e-mail to WG list).

Two I-Ds moving forward on trunking and migration from Adamson and Noveck, again see slides. In the next month. Adds trunking as an fs_locations attribute feature. Will update 7530, not 7931. See comments on Adamson trunking update draft in slides. Client multipath document will be allowed to expire.

-- Flex-File update (Haynes) (20 minutes)

flex_files went through last call some time back, but review comments came in late and they have to be addressed.

Haynes decided that we have a model for treating the state-Id to avoid breaking existing implementations.

See slides on resolving the MDS and KDC ids.

WGLC will be pushed through next. (get the other document for GLC from Haynes - put doc names in

-- Next steps for NFS RDMA (Lever) (25 minutes)

Chuck described performance benefits seen with RDMA.

Chuck discussed future enhancements to NFS RDMA, for a potential new version. Integrity and confidentiality of transport headers for ROCE bears investigation.

-- NVMe and NFS (Hellwig) (10 minutes) 

Described opportunities around new'ish technology.

David Black joined and added comments on SCSI reservations.

-- RDMA Layout Type (Hellwig) (20 minutes)

For NVME. Roughly Christoph wants to recast SCSI layouts for use in NVME as data stores.

-- Where To Next (Trond (missing) and Haynes)

Scale out and replication. Using pNFS for replication. 

vsock is a new VM to VM protocol, picked up by Linux. It's not documented in its use of NFS.

David Black wants some discussion between vsock people to talk to container folks (shutdown and restart problems are related to live migration).

Chuck says we need to identify the things that do not work with containers etc. today. Follow up with owners of Kubernetes etc.

READDIR statelessness and persistent cookies are a pain - but no other solution.

-- Recharter discussion (Noveck) (20 minutes)

We've been doing things outside our current (dated) charter.

Security considerations. Maybe need a 7530 security considerations doc. Model on some transport area approach to context.

Spencer Dawkins suggested a simplification of recharter wording. As did David Black. But the wording that Noveck came up with seemed acceptable and reflected the working group activity and goals.