OpenZFS Developer Summit 2013

The first OpenZFS Developer Summit was held November 18-19, 2013. The goals of the event were:

• to foster cross-platform community discussions of OpenZFS work
• to make progress on some of the projects proposed for this community.

Video recordings of the presentations are in the OpenZFS channel on YouTube.

Topics that were discussed

Title	Speaker	Slides	Video
Introduction	Matt Ahrens	slides	video
Platform Panel	Brian Behlendorf, Jorgen Lundman, Chris Siden, Xin Li	-	video
Platform-independent code repository	Matt Ahrens	slides	video
Storage Tiering	Boris Protopopov	slides	video
Vendor Lightning Talks	representatives from each participating company	-	video
Community Planning	Karyn Ritter	slides	video
Channel Programs	Chris Siden, Max Grossman	slides	video
Test Coverage	John Kennedy	slides	video
Performance Investigation	Adam Leventhal	slides	video
Performance on full & fragmented pools	George Wilson	slides	video
Scalability	Kirill Davydychev	-	video
Virtual Memory Interactions	Brian Behlendorf	-	video
Multi Tenancy	Rob Mustacchi	-	video
Examining On-disk Format	Max Bruning	-	video

Hackathon projects

These projects were worked on at the hackathon:

• Testrunner suite running on other platforms: linux, freebsd (+ illumos libzpool)
  • John Kennedy, Steven Burgess, Ned Bass, Tony Nguyen, Kirill Davydychev, Matt Ahrens
  • Got several tests running on Linux
    • "it's a giant hack" -- lots of work needed to get this to integration quality
  • Got several tests running against libzpool on illumos (using userland ioctl shim)
• channel programs: pass instruction stream into kernel
  • Chris Siden, Max Grossman, Alex Reece
  • See project page, slides and video describing project
  • Got snapshotting via a Lua program working all the way through the stack
• ZFS send progress reporting with "set proc title"
  • Xin Li
  • Finished and sent for code review on FreeBSD
  • "ps" and "top" can display zfs send progress: "sending pool/fs@snap X MB done (24%)"
  • small amount of work needed to hook into title-updating routines on Linux and illumos
• dedicated kmem_cache for dedup table entries
  • John Layman, Matt Youds, Kirill
  • goals: increase diagnosability, reduce memory consumption
  • Finished this and also added kmem_cache for l2arc header
• larger (1MB) blocksize
  • Brian Behlendorf
  • Got it working, can set recordsize property to 1MB!
  • Next steps: performance investigation, thorough review of all uses of SPA_MAXBLOCKSIZE
• compressed ARC
  • George Wilson, Rob Mustacchi
  • got it working! (modulo some refcount bugs)
  • Next steps: figure out better policy for when to compress/uncompress
• upstreaming dbuf_user & dbuf state machine (spectra DMU changes)
  • Justin Gibbs, Tony Nguyen, Boris Protopopov
  • Lots of progress, got it compiling against latest illumos code and passes several tests
• limit DDT size
  • Sanjeev Bagewadi, Prashanth Sreenivasa
  • got it sort of working w/ limit set by global tunable
• onexit for osx
  • Jorgen Lundman
  • pretty annoying to implement since OSX doesn't give much visibility into file handles, vnodes, or device handles

Event Planning Info

See OpenZFS Developer Summit 2013 planning

A huge THANK YOU to the sponsors of the 2013 Developer Summit: