LWG Minutes 2014-10-01

Agenda

 * Contract Updates
 * Roadmap Update
 * Lustre's Future

Attendance

 * Chris Morrone (LLNL)
 * Andreas Dilger (Intel)
 * Eric Barton (Intel)
 * James Simmons (ORNL)
 * Peter Bojanic (Seagate)
 * Justin Miller (IU)
 * Peter Jones (Intel)
 * Richard Henwood (Intel)
 * Cory Spitz (Cray)
 * Ruth Klundt (Sandia)
 * Steve Young (Seagate)
 * Colin Faber (Seagate)

Contract Updates
SFS-DEV-004 (CLIO Simplification Implementation) is signed and about to begin work. Final call for PAC participants given. Three people on the call volunteered. Contract wiki page is at:

http://wiki.opensfs.org/Contract_SFS-DEV-004

The Lustre Protocol Documentation project is under negotiation with Intel. A call for "pre-PAC" volunteers was communicated. The early PAC members will help Chris with the contract work.

Lustre's Future (and OpenSFS's role)
Peter Bojanic revived discussion about Lustre testing from previous meetings. He made the suggestion that robust acceptance tests are more valueable than documentation.

Chris suggested that, while testing is extremely valuable, documentation is arguably more valueable than testing because it informs all stages of the software development process including the tests that need to be created.

Chris talked a bit a possible direction for OpenSFS to focus its effort: spend most new funds on tackling the long standing Lustre technical debt directly, mostly in the forms of documentation and code refactoring.

Chris suggests that our current approach of trying to handle technical debt as a side effect of feature development is not dealing with the problem at an accepatable rate. Technical debt is also an area that individual vendors and other organizations are unlikely to tackle on their own. He suggests that OpenSFS should be the place where we can come together to take this necessary and challenging work.

Peter Bojanic agreed that sales forces at vendor organizations tend to prioritize features over foundational work.

Eric Barton wondered if it would be a good idea to address the design defficiencies in Lustre that make it less than tolerant of various forms of system failures.

Chris thought that it would be very nice to do that, but making further changes to the Lustre protocol without having a documented protocol is asking for trouble for several reasons, including that the Lustre client code is now in the staging area of the Linux kernel. Maintaining protocol compatibility is already a weak point for Lustre, and is only going to get harder without protocol documentation.

Action Items

 * Chris to make mailing list for SFS-DEV-004 PAC members and related Intel personnel
 * Continue work on Lustre Protocol Documentation Project
 * Make wiki page listing Lustre Reviewers by topic area