[Twg] OpenSFS TWG meeting notes

[David Dillow]

Next meeting:
	TBD, tentatively November 7th, 2013

Participants:
	Cory Spitz		Cray
	Alex Kulyavtsev		Fermi
	Doug Oucharek		Intel
	Bryon Neitzel		Intel
	Andreas Dilger		Intel
	Steve Simms		U
	Andrew Korty		IU
	Josh Walgenbach		IU
	Tom Crowe		IU
	John Fitzpatrick	MWR Infosecurity
	David Dillow		ORNL

Agenda:
      * Farewell and thanks to John Carrier for his work with OpenSFS
      * Plans for TWG going forward
              * Time to restart requirements gathering process
                      * more input from vendors' sales teams; they know
                        some of the pressure points the technical folks
                        do not, such as why Lustre may loose out to our
                        competitors in a bid
              * Monthly technical presentations
              * Need to find more outside presentations
              * Happy to include talks from OpenSFS projects, but they
                only take us so far and there is great work in the
                community that deserves to be highlighted
      * Status of the Intel contract (DNE/LFSCK)
              * Artifacts stored Intel HPDD wiki, and on openSFS wiki
                      * http://wiki.opensfs.org/SFS-DEV-001_status
              * DNE 2: Striped Directories
                      * Implementation in progress, expect completion
                        mid-November
              * LFSCK 2: MDT-OST consistency
                      * Implementation near completion, running a little
                        behind, but should be complete in time for 2.6
              * Status of the new development contract with Intel
                      * Steve Simms is technical representative
                              * Shared PAC members with first Intel NRE
                                contract
                              * Shared meeting time, open to outside
                                attendees
                                      * Send email to ssimms at iu.edu for
                                        details
                      * Design projects to sharpen plan and estimates
                        for implementation
                              * Solution architecture and high-level
                                design as deliverables
                              * Expect to have accurate estimate of
                                effort for implementation for Board
                                decision to proceed
                      * Thrust areas
                              * Layout enhancement (Design)
                                      * N-way stripe replication
                                      * erasure codes
                                      * layout extents (e.g. change as
                                        file grows)
                              * CLIO simplification (Design)
                                      * Added platform portability
                                      * Performance significantly lower
                                        than 1.8
                                      * Simply cl_lock (going cacheless)
                                      * Removing vvp and lovsub
                                        portability layers
                                      * Remove obsolete/dead code
                                              * obsolete ODB callbacks
                                                (brw, punch)
                                              * liblustre
                              * Data on MDS (Design)
                                      * Allow small files on MDT for
                                        performance
                              * Dead Code Removal (Implementation)
                              * Code Documentation (implementation)
                                      * Add commenting best practices
                                      * LOD, OSP, OFD
                                              * most recent rework,
                                                lacking clear docs
      * Deep dive into the IU uid-mapping and shared key work
              * Joshua Walgenbach on UID/GID mapping work
              * Andrew Korty on challenges implementing Shared Keys
                using GSS
      * Q&A
              * How goes Lustre in the mainline kernel?
                      * Good initial push, but needs ongoing attention
-- 
Dave Dillow
National Center for Computational Science
Oak Ridge National Laboratory
(865) 241-6602 office