NFS version 4 - Maastricht, The Netherlands Wednesday, July 28, 2010 13:00 - 15:30 0.1 London Conf Room ------------------------ - Intro/Blue Sheets/Note Well (Pawlowski) Two Microsoft IP disclosures. https://datatracker.ietf.org/ipr/1361/ https://datatracker.ietf.org/ipr/1362/ Refer to the links on IPR disclosure in the Note Well: http://www.ietf.org/about/note-well.html And consult legal departments perhaps on potential impact of disclosure license terms if in effect. Decade working group - Rich Bo? Requirements document. Separate the storsge requirements from the App requirement, peer-to-peer caching, etc. Looking for technical advisor... Decade WG may have another use case for this policy-based approach... Per file metadata - longevity is example. Peer-to-peer examples of limited duration. === NFSv4 Migration Issues - Trond Myklebust - 10 minutes Solaris and Linux kernel, client migration issues. Aren't documented well in 3530. How you are supposed to recover after a migration? Recap of issues. Errata to enter 3530bis - Tom Haynes taking as 4.0 issue. Permanent session is one way to get around it? === pNFS Access Permissions Check - Sorin Faibish - 15 minutes draft-faibish-nfsv4-pnfs-access-permissions-check-03 Add two additional layout return types. Put error code in part to cover any loss of data server connectivity. See slides. XDR fragments - minor changes, require a later minor version. Can be done in 4.1 - two new error types... Unofficial minor version extension. Next draft coming as a -00 to working group. Mandatory vs. Optional error codes... A lot of discussion on frequency/reality of needing to cover this error scenario. Sorin said current (non pNFS) product has seen these errors, as has Panasas product (Halevy). Concern with pNFS over objects - Halevy - wants one way to return errors... David Black and Benny to work offline on this. Transient errors - report it - Eisler, error code and byte range... === pNFS Storage Device Preference - Faibish/Black/Eisler - 20 minutes Goal identifying usage and requirements (policy) for the file instead of specific placement information. Came from pNFS - will work with NFS version 4.1, separation of name, quality of service... Portability of hints - MPI. Extend POSIX - this has been done before. Per implementation optimizations. Dean observes this never cleared the ground before - and may be infeasible. WheelFS - MIT research paper. Pointer to be sent to mail list. Solution details. Likely XDR, to pass values. Earlier David Black suggested concern of metadata overhead for small files - which I think was countered by inheritance from directory of default value. -00 draft to follow. David Black is going to point to slides on list and ask for questions before Beijing. === Sparse and Direct I/O Support Simple and Efficient Read Support for Sparse Files - Hildebrand - 10 minutes draft-hildebrand-nfsv4-read-sparse-00 Any application that uses sparse files... reading the holes suck from a client perspective. Suggested change to protocol. For 4.2... Corresponding support in SCSI. Also done in OSD SCSI protocol. Direct I/O separate discussion. Large block size writes for direct push. 4.2 proposal to add fadvise() POSIX style. Will help with O_DIRECT. Shepler to talk to Dean on how to move forward. === FedFS I-D progress/issue identification - Craig Everhart - 10 minutes Spencer to move draft forward - no WG action needed (see slides). I18N - will follow NFS 4 lead on naming. Shepler to shepherd forward before Oct 2010. NSDB protocol. Chuck Lever at Oracle is constructing an admin server. Open question on admin protocol - throw something to alias on Craig's issue. === NFS Version 4 Protocol - Haynes - 10 minutes draft-ietf-nfsv4-rfc3530bis-04 draft-ietf-nfsv4-rfc3530bis-dot-x-03 There is a subversive directory for the issues, ticket system and tracking list. Sep 2009 was the milestone... RFC Editor errata (1) against this. WHEN? Spencer, David Black, Tom Haynes, etc. Triage items. === i18n update to 3530bis - Noveck - 10 minutes See slides... File Open is at stake here - you pass in a file name that is compared to file name in FS underlying server. If either it fails to find a file that you once created, or open an incorrect file.... David Black. Two things: Spencer wants to drive via use cases. And David Black maybe to briefly summarize to working group alias of related activity in the IETF. === 4.1 client - LAYOUTCOMMIT & close - Myklebust - 10 minutes See slides. Questions around V2 and V3 data servers? Issue around client failing to do a layout commit. Summary back to list of problem understanding and drive solution either in change or wording change... errata? The question is: what it is. === NFSv4.2 - 20 minutes Out of time... === Wrapup (All) Out of time...