Last week in San Francisco the Linux developer community met to collaborate on and discuss the future of Linux. I attended, and tried to glean the most important developments that will be affecting Linux server admins and users.
1) Control groups (cgroups): Cgroups was initially developed to limit resource usage in the Linux kernel. Memory, bandwidth, CPU usage can be controlled using cgroups, and it can be used to deny access to and monitor system resources. Red Hat’s Resource Management Guide has a great description with a lot of detail on how they work, check it out for more info. The developers are still working on improving how cgroups work, so report bugs and stay tuned!
2) KVM: It’s really the hypervisor that will be most supported by developers… so get ready to switch from Xen if you haven’t already. Christoph Hellwig gave a thorough talk on how KVM and qemu handle presenting local storage to the guest and what developers are doing in that area. Mike Day of IBM set out to debunk some myths about KVM that have held it up from being adopted more broadly. Some of his points were well-received by the audience, but there was a little dissension in the audience when he tried to claim that KVM and VMware were very similar because VMware’s VMKernel is largely based on the same Linux code… Audience members (who work at VMware) said no it’s not. It hasn’t been for a while.
3) Yocto Project: While you may not care much about embedded devices, who knows how the world (or your job) might change in the future? Besides, Linux is in everything if it’s electronic. The Yocto Project is the combined effort of all the major embedded chip vendors, embedded commercial Linux vendors, individual developers and OpenEmbedded to help developers not have to reinvent the wheel each time they go to create a new device. Check it out.
4) File systems: A great talk from Michael Rubin of Google provided details on why Google chose ext4 to deploy to replace ext2 (and a little about their process of doing it slowly so they didn’t lose all the data). XFS and ext3 were considered but dismissed, the former due to its complexity, the latter because it contained some of the same drawbacks as ext2. While Google chose ext4, Rubin seemed to have caught the buzz around btrfs, which just wasn’t “mature” at the time. With a three year transition from ext2 to ext4 though, I’m doubting Google will be looking to make another move any time soon. But if you’re looking at a file system upgrade, it may be worth looking into “butter.”
5) Open vSwitch: Open vSwitch is a network switch built for virtual environments and differs from traditional switches by exporting an external interface for fine-grained control of configuration state and forwarding behavior. There was a well-attended presentation on this technology and it’s certainly something to watch or look into if you are running large virtual environments and need more efficiency.