Kernel Log: Ceph file system in 2.6.34, kernel and KVM presentations at the CLT2010
by Thorsten Leemhuis
Linus Torvalds has integrated the Ceph distributed network file system and released the second release candidate of 2.6.34. Torvalds intends to vary the length of the merge window from now on. Slides of the presentations at the Chemnitz Linux Days provide background information about kernel and KVM topics. Ubuntu has been given an AMD graphics driver which cooperates with series 1.7 X Servers.
When he released the first RC of 2.6.34 early, which surprised many subsystem maintainers, Linus Torvalds announced that he might still integrate the Ceph distributed network file system, which didn't make it into 2.6.33, although the merge window had already closed. He did this last weekend and, shortly afterwards, released Linux 2.6.24-rc2 – the usual release email for the new version has not so far appeared on the LKML.
Licensed under the LGPL, Ceph is a "Distributed Network File System" which, according to the developers, can manage data on a petabyte scale "and beyond", it is robust and offers numerous functions that are missing in other similar open source file systems. Detailed information about Ceph is available on the project's homepage, in a short description which was integrated into the kernel sources together with the file system code, and in an older article on LWN.net which describes an earlier, Fuse-based version of the file system.
The file system wasn't the only major addition Torvalds made after closing the latest merge window two days earlier than usual. Between rc1 and rc2, multiple changes were also made to the Btrfs file system, the SCSI subsystem and the architecture code for ARM, Blackfin and Microblaze CPUs – more than are usually made at this stage, which generally only accommodates a few stragglers. Several days earlier, Torvalds had actually refused to integrate the SCSI changes in order to teach the maintainers of the SCSI code and other subsystem maintainers a lesson. Torvalds hoped to prompt them to stop submitting their changes at the last minute, however, it appears that in the end, he did show mercy.
Torvalds has clarified that he now intends to vary the length of future merge windows – the period of time in the development cycle in which the majority of changes are integrated into a new kernel version. He said that he won't mention in advance how long the respective merge window is going to be – and that he might close it after seven days if he thinks that there is enough work done. ("[...]I'm not even going to mention in the release notes how long the merge window is going to be. Maybe I'll say "that's enough" after just one week, [...]"). With these measures, Torvalds has further adapted the (partially unwritten) rules of the Linux kernel development process.
In one of the auditoriums the majority of the presentations during the recent Chemnitz Linux Days discussed the Linux kernel and its KVM hypervisor. Some of the, mostly German-language, presentation slides and short papers provide ample information for all those who couldn't attend the event:
- For instance, Robert Richter's presentation "Git is MacGyver - managing kernel sources with Git" (Slides) explained numerous practical aspects of using Git in kernel development.
- In "Einführung in das Linux Memory Management (Introduction into Linux memory management)" (Slides), Johannes Weiner provides some background information about the Linux kernel's memory management.
- Stefan Assmann's presentation "Einführung in Real-Time Linux (Introduction into real-time Linux)" (Slides, short paper) explains the significance of real time (RT) and demonstrates how to build a custom real-time kernel.
- Bernhard Walle's presentation on "Kernel-Debugging (Kernel debugging)" is not just available on Slides and as a short paper, there is also a text version.
- Those who regularly read the Kernel Log at The H Open are unlikely to find many surprises on the slides(direct download) of the presentation entitled "Aktuelle Entwicklungen beim Linux-Kernel (Current developments around the Linux kernel)", because this presentation was given by the author of the Kernel Log and generally covered topics which have already been discussed.
- The slides of Jörg Rödel's presentation entitled "Patch Applied! - Working with the Linux kernel community" offers some important information for all those who want to contribute to the Linux kernel.
- Perhaps rarely significant for desktop systems, but crucial in High Performance Computing (HPC) is the distribution of processes across different cores and processors in multi-processor systems, which is the topic discussed by Andreas Herrmann in "Why CPU Topology Matters" (Slides).
- Alexander Graf gave two presentations about KVM: "KVM on PowerPC" (Slides) and "Nesting the Virtualized World" (Slides).
- In "Live and in Colour - Live Migration?" (Slides), André Przywara presents some background information on migrating virtual machines between systems without having to suspend the guest system or the server processes running on it.
Some of the many volunteers who organised the Chemnitz Linux Days are currently working to add video and audio recordings of the presentations in the five main auditoriums to the event's homepage. The organisers of the linux.conf.au (LCA) 2010 held in Wellington, New Zealand, have recently completed this task and have released videos of many of the presentations. The kernel and its environment were also a topic there – for example in the "Kernel Report" by LWN.net founder and figurehead Jonathan Corbet. The slides of many LCA 2010 presentations have been available via the conference wiki for some time.
- Greg Kroah-Hartman has released the stable kernel versions 126.96.36.199 and 188.8.131.52 which, as usual, offer several dozen corrections and minor improvements; the release email contains the traditional recommendation that users update to one of these versions.
- Linus Torvalds refused to integrate the current form of "writeable limits", which are intended for a future version of SLES and were submitted by Jiri Slaby for integration into 2.6.34.
- Nigel Cunningham has released version 3.1 of TuxOnIce – an alternative implementation for using software suspend (suspend/hibernate). At the end of his release email, Cunningham indicates that he intends to focus on improving the kernel's software suspend code once he has completed some remaining TuxOnIce-related tasks.
Kernel environment ("plumbing layer") and userland drivers
- Kernel.org administrator John 'Warthog9' Hawley has announced that Kernel.org-hosted services such as Bugzilla, Wiki and Patchwork are now available via SSL.
- Greg Kroah-Hartman has released version 0.87 of the usbutils.
- Neil Brown has released version 3.1.2 of mdadm, whose default metadata format for software RAIDs created via the kernel's MD code is now v1.2.
- In a LWN.net article which is now also available to non-subscribers, the Linux Weekly News team have compiled numerous aspects of the interaction between Linux and storage media with 4-KByte sectors. This topic has also been repeatedly discussed in the Kernel Log. Furthermore, it recently sparked prolonged discussions on the LKML; the page in the Linux ATA wiki used by kernel hackers for the relevant background information has recently been given another major overhaul; corrections were, for instance, made in some areas that discuss running a Linux distribution in parallel with older Windows versions such as XP.
- Peter Hutterer has released version 1.7.6 of X.org's X Server, which only introduces minor corrections.
- The latest proprietary Linux graphics drivers on AMD's driver web page don't work with the series 1.7 X Servers that have been available since last October. However, the Ubuntu developers apparently received the pre-release version of a driver that fixes the problem for Ubuntu 10.04, which includes such an X Server. As this driver is not publicly available at AMD, this doesn't really help the users of other distributions with X.org 7.5 X Servers. Ubuntu has received such special treatment for all versions since Ubuntu 8.10, while AMD has, at times, left the users of other distributions out in the cold for months.
- The beta version of a proprietary NVIDIA driver for GeForce graphics hardware already supports OpenGL 3.3.
- The developers of the ATI X.org driver package have released versions 6.12.6 and 6.12.192 of their drivers for Radeon graphics cards; the latter version is the second release candidate of the series 6.13 drivers and is said to considerably improve the performance of KMS EXA DFS (and, therefore, of GetImage).
- Michel Dänzer has released version 11.0.1 of the X.org graphics drivers for VMware guests.
- On the X.org developer mailing list, Peter Hutterer has provided an overview of what he thinks is required for X to provide decent multi-touch support.
- In his blog, Dave Airlie describes some experiments that involve running two X Servers on one graphics card – this is interesting for scenarios such as multi-seat environments.
- In a second post, Airlie explained some aspects of a "proof of concept" in which a Radeon graphics card computes images that are subsequently output via Intel chip-set graphics.
Older Kernel Logs can be found in the archives or by using the search function at The H Open Source. New editions of Kernel Logs are also mentioned on Identi.ca and Twitter via "@kernellog2". The Kernel Log author also posts updates about various topics on Identi.ca and Twitter via "@kernellogauthor".