NFS version 4 - Ville de Quebec (IETF 81) Monday, July 25, 2011 0900 - 1130 : Room 2103 -------------------------------- === Intro / Blue Sheets Noted Well etc. === Agenda bashing Many additions; final content for Monday follows - Agenda bashing - IPR disclosure (beepy) - Shepherding update on I-Ds in pipeline and coordination with AD (Shepler) - Charter, milestones, open discussion (Shepler) - RFC3530bis Last Call review and closure (Haynes) - LAYOUT_COMMIT errata discussion and resolution (Black) - RPCSEC_GSSv3 (Haynes) - NFSv4.2 (Haynes) - Labeled NFS (Haynes) - Layoutget during cache writeback (Sorin) - I18N impacts on NFSv4(x), FedFS (Black) - Object Storage (Eisler) === IPR disclosure (Beepy) US 7,933,921 April 26, 2011, Referent-controlled location resolution of resources in a federated distributed system. Royalty-Free, Reasonable and Non-Discriminatory License to All Implementers. License to be submitted to IETF per usual process. === Shepherding update on I-Ds in pipeline and coordination with AD (Shepler) FedFS - three documents complete. Review and updates, waiting on final write-up to push to DBH (Harrington). DNSserv may come up as an issue in wider review as appropriate or deprecated. Shepler thinks this will go through okay, but will give heads up in the shepherding document. Shepler thinks we have cleaned up, but the I18N considerations may cause an edit. Submit documents and work in parallel. Apps Area Directors. Spencer has some text for the shepherding documents, three to go forward at end of week. Shepherding of Tom's 3530bis following resolution of latest issues and notification to working group. If WG last call needed it will be short. === Charter, milestones, open discussion (Shepler) Harrington: focus on delivering Are we getting closer to the end, or further from the beginning. ??? Black: charter bash I18N.next, Thursday afternoon. Take lid off that Pandora's box. Anything that uses text. File names are like nothing else in the IETF - we need to look at this. [Who will be I18N lead in NFS V4]. Charter update keep 4.0 and 4.1 and RPC/XDR in maintenance. Move FedFS to maintenance. Today's discussion on proposed milestones for bis document today. Only addition for the charter. Only open doc is RPCSECgss V3, Spencer has been dealing with it as RPC maintenance. Shepler and Benny Halevy - waiting on direction from Panasas, RFC 5664bis will be submitted (errata has been submitted required for draft). Accept errata as they are, and still do the bis work. Benny suggests submitting a refresh of the text as extensive, but no change. Benny will do the bis work. === RFC3530bis Last Call review and closure (Haynes) Tom Haynes. On schedule for the late schedule. Entered and left WG last call. Problem found in last call, 3530bis, lockers issue. Minor typos. No last call. Map a section of changes for him to review - help DBH go through AD review. Spencer asks if we need directed review guidelines for the bis document, and then who will write it. Review of the draft will take a long time. Will be submitted with a general shepherding document? Shepler will follow up with a bit more detailed change description as aid to subsequent reviewers. Three weeks from today to deliver bis document. Any RFC 2119 requirements? Brief One Week Last Call to address those if necessary. === LAYOUT_COMMIT errata discussion and resolution (goal for final text/consensus) - implications/impact for NFSv4.2 David Black. LAYOUT_COMMIT is operation to cause the client to commit data to a data server. The start and end of layout parameters have turned out to be problematic for block layout. File and Object always commit everything. In block version, layout has blocks changed approach, and start and end not needed. In Prague decided to eliminate, the question is how. Two choices: sent the range to (0,0) or range of entire file, and server to ignore range. Will there be a problem with server behavior of RFC? David Black can go either way, and write 4.2 text. The hard part, is doing something for the existing RFC. MUCH DISCUSSION. Some discussion on performance, may be or may be not. Will consider in 4.2 write-up. === RPCSECgss V3 (Haynes) Can used RPCSEC without Kerberos, Nico. Need at least two more rounds of edit/review. Nico has a set of changes waiting for EMC feedback. EMC gave first round of feedback, beyond that prepared to live with it as spec'ed - David Black. Problem was procedure numbers being out of order, David Black to send clarifying e-mail. Sorin - we don't want to force people to implement gssV3 who don't need it. Related to the server-side copy (clarified in the draft already - David Black). The situation with server side copy, to get the security completely right, you need to get the delegation right (server acting as the user). David Black has been writing up with Tom Haynes how to do server side copy without gss V3. Server side copy is an optimization. Shepler to work with Tom and Nico to do the work. Make sure you get the security requirements straight lands in the 4.2 draft - David Black. [Later] In response to Sorin, Nico says RPCSECv2 should simply be obsoleted. === NFSv4.2 (Haynes) - review of design group work since last IETF meeting- schedule/plan for future conference calls READ_PLUS - consensus on open issues on this piece. Error code to differentiate unsupported union portion, instead of only opcode. Labeled NFS is covered in its own meeting. No controversy about space reservation. === Labeled NFS (Haynes) - Significant progress has been made - Discussion needed of resulting content and assurance that consensus exists given current I-D Minimal XDR changes. See Tom's slides. Jared is happy. === Layoutget during cache writeback (Sorin) Clarification needed on expected behavior, Sorin to take to working group e-mail to drive consensus. === I18N impacts on NFSv4(x), FedFS (Black) Covered before. Input from emerging IETF considerations about I18N, in Precis. Figure out if anything to back patch, then back patch it. === Object Storage (Eisler) (10 minutes) Not for charter addition. Nameless object fetching. Recent APIs in Linux can support this. Reduce overhead of traversing directories through file handle only access. Benny adds OSD security model. === RPCSEC_GSSv3 - Review of dependencies on this work from NFSv4.2 (e.g. server-side copy) NFS version 4 - Ville de Quebec (IETF 81) Wednesday, July 27, 2011 1510 - 1610: Room 2103 -------------------------------- === NFSv4 Fadvise (hildebrand) (20 minutes) draft-hildebrand-nfsv4-fadvise-02.txt Dean and Mike had met prior to the meeting and have resolved the high-level pieces of their proposals. Working towards a joint I-D that will merge the approaches into a single method. Dean's presentation material covers the -01 draft submitted 2-3 weeks ago. The approach is derived from Posix fadvise() in an attempt to capture the behavior such the server can provide, if supported, the hints. Two main comments during the discussion. While the desire is to have the new operation (IO_ADVISE) preceed READ/WRITE operations and not block a compound, the issue is that new operations, by default, can not be mandatory to implement. Need to resolve this if the approach is to be taken. Second method of combining hints, if at all possible, is important to clarify the semantics of what the client and server can expect of each other. === I/O hints (Eisler) (10 minutes) Bulk of discussion focused on complexity of some of the proposed hints and whether the functionality fit into the NFSv4.2 timeframe. Specifically, the multi-level cache hints were the most discussed. Plan is to combine drafts and then hold a series of 2/3 conference calls to work out remaining details. === pNFS Block LUN Identification Sorin reviewed the need of GUID identification of pNFS LUNs in an effort to avoid OS mis-identification and inappropriate use. The current draft suggests a GUID value. Another update is needed to clarify the requirements of GUID identification and the implication for other bits of OS software. Sorin also suggested another use case and a following I-D to capture those problems. Will not disrupt the current I-D.