Xen Maintainer, Committer and Developer Meeting/January 2013 Minutes

From Xen
Jump to navigationJump to search

Minutes

Attendees

  • Ian Campbell
  • Lars Kurth
  • Andres Lagar-Cavilla
  • Jing Su
  • Daniel De Graff
  • Daniel Kiper
  • Boris Ostrovsky
  • Olaf Herring
  • Sherry Hurwitz
  • James Bulpin
  • Jan Beulich
  • Jason Douglas
  • Konrad Rzeszutek Wilk
  • Don Dugger
  • Jun Nakajima
  • Matt Wilson

Review of Action Items

  • Konrad to talk about blk protocol improvements on xen-devel (DONE)
    • Some discussion on list, progress being made.
  • Intel/Jan to send a patch against MAINTAINERS for generic IOMMU parts (DONE)
  • AMD to nominate a new maintainer for the AMD code. IN PROGRESS
  • Ian to send out new dial in details to everyone (DONE)

Current technical challenges

  • ACPI ERST handling (Jan)
    • Jan: There was a patch before 4.2 to correct ERSE table handling but it caused regressions in the test system and was reverted.
    • The current code only accepts invalid tables from early SDVs and rejects correct tables!
    • However fixing this causes the correct tables in some production machines to trigger a hypervisor crash.
      • Only known such machines are in the Xen.org osstest pool
      • Both SuSE (Jan & Jason) and AMD (Boris) have tried (unsuccessfully) to find a machine which can repro.

    • Pictogram voting comment 15px.png Action Ian C: to check with Ian J that beetle machine have up to date BIOSes

    • Pictogram voting comment 15px.png Action Boris: to try again to find a suitable machine.
    • As a fallback we may be able to arrange for external access to the osstest pool.

Coordination on current and future work

  • Performance measurement.
    • Sherry asked what strategies people were persuing WRT performance analysis tools
    • Konrad: Can divide into oprofile working, vs. oprofile upstreamable
  * vs. working on perf.
      • oprofile works for PV but not for HVM (unsure why)
      • Konrad thinks implementing perf support is a good idea, was a GSOC project but no one took it up
      • Sherry: Should AMD look into this?
        • Konrad: Oracle currently doesn't have resource
        • IanC: Citrix also doesn't
    • Jun: Intel have been working on virtual PMU and maching pchan work in HVM guests .
    • Jan: Need for a tool to do global performance monitoring, including the hypervisor and not just within guests.
    • Don: Tools for end users are also important, e.g. for measuring applications on virt. vs. native
    • Matt: Hypervisor and/or global profiling is more important to them.

    • Pictogram voting comment 15px.png Action Konrad: to start a conversation on xen-devel

    • Pictogram voting comment 15px.png Action James: to ping XenServer performance team to see if there is someone who should partcipiate.
  • KEXEC
    • Daniel has posted patches to implement kexec in Linux and they were not accepted
    • New approach is an improved hypercall interface which will allow this to be implemented in userspace
    • David Vrabel (XenServer) is also looking at some kexec related development

    • Pictogram voting comment 15px.png Action James: to ping David to post on xen-devel and coordinate with Daniel.
    • Jun asked why kexec on Xen?
      • Konrad: priority is to kexec on crash, secondary desire is to kexec within PV guests.
      • Daniel: Kexec already works in an HVM guest.
      • James: XenServer is also interested in dom0 crash dump
  • Testing for security
    • Jun asked what security testing people do. Is concerned that as more complexity is added simple smoke testing may not be sufficient?
    • Ian: The osstest system has been made more consumable and it should be possible for people to write new tests against it now if people are interested in adding security relevant test cases
    • James suggested Mirage for use as a test VM.

    • Pictogram voting comment 15px.png Action Lars: to ping Ian J regarding GSoC projects on the test system (DONE)

Community news, activities

  • Code contributions Xen vs. KVM
    • Lars has run some stats
    • Hard to determine what is "Xen" and what is "KVM" (e.g. pvops, qemu, tools etc)
    • Within the constaints of this it seems the two communities are similar in terms of numbers of contributions, churn, breadth of community, numbers of organisations etc etc

    • Pictogram voting comment 15px.png Action Lars: to circulate the final 2012 analysis.