Linus Torvalds announced the release of a stable (Linux 2.6.32 kernel. Major additions include kernel-based mode setting (KMS) and 3D graphics support on select Radeon cards, plus new kernel shared memory (KSM) technology for KVM virtualization, power-saving and performance improvements, and a faster “Devtmpfs” boot technology.
Many of the recent Linux kernel releases have focused on filesystems, such as September’sÂ Linux 2.6.31, which was notable primarily for adding USB 3.0 support, but also offered a much more robust implementation of the Btrfs filesystem. Before that,Â Linux 2.6.30added the NILFS2, EXOFS, and POHMELFS filesystems, andÂ Linux 2.6.29 debuted the embedded-oriented Squashfs filesystem, as well as the first implementation of Btrfs.
With Linux 2.6.32, however, “I think we have a release without any actual new filesystem for once,” writes Torvalds (pictured at top) in the announcement. “But there’s been a fair amount of changes to btrfs, and the block layer writeback itself has been through major updates, and the whole per-bdi writeback thing is a pretty big change. Other rather noticeable changes are in . . . drivers all over the place.”
Over atÂ The H, Thorsten Leemhuis offers a thorough rundown on the new kernel features, which are led by the support of KMS and 3D graphics on AMD/ATI’s series 2000, 3000 and 4000 Radeon graphics cards. Improving 3D graphics support should help improve Linux’ standing as a gaming platform, among other benefits.
Meanwhile, performing mode-setting for graphics card resolution and depth mode settings in kernel-space (KMS) offers advantages over user-based mode-setting, such as increased flexibility and security in case of a fatal error in the kernel. As welcome as these developments are in bringing Linux up to snuff for graphics, Leemhius notes that ATI just released Radeon series 5000 graphics cards, which are supported only by generic VESA open source driver.
KSM improves virtualization performance
Not to be confused with KMS is the new kernel shared memory (KSM) technology, which “merges identical memory pages from different userland processes to reduce memory usage in virtualized environments and make more efficient use ofÂ hardware,” writes Leemhuis. Also referred to as “Kernel SamePage Merging,” KSM derives from the KVM virtualization community. KSM is said to scan the memory of multiple userland processes for identical areas, and then combines any matches, improving performance by reducing the memory load. This is particularly useful when “multiple similar guest operating systems share the same software libraries and programs run on a computer, causing large areas of the data stored in the guests’ memory to be identical,” he adds.
With improvements such as KSM over recent kernel releases, KVM has been improved to the point that it “has prompted Linux heavyweight Red Hat to shift the focus away from Xen, which was all the rage in the field of virtualization a few years ago,” writes Leemhuis.
Devtmpfs promises faster kernel boots
The somewhatÂ controversial “Devtmpfs” boot system, meanwhile, was said to be favored by Torvalds. Devtmpfs “should mean that the Linux kernel boots faster and no longer requires udev, while new make targets will allow testers to easily generate kernel configurations adapted to their systems,” writes Leemhuis.
The “per-bdi writeback thing” mentioned by Torvalds is a “major rewrite of the writeback infrastructure,” according to Leemhuis, resulting in each device now being linked to its own “per-BDI” thread. Along with other improvements, the writeback overhaul “should significantly increase data throughput for writeback-intensive access scenarios and cause them to run more evenly,” he adds.
Other improvements include support for Intel’s upcoming Moorestown processor platform, which is aimed primarily at MIDs, and major changes to the power management code. Meanwhile, performance has been improved in a variety of ways, including enhancements to the Cpuidle framework, the block layer, and Btrfs, according to Leemhuis.
“Linux 2.6.32 should not only allow PCs to push more data around, it should also feel faster, with changes to the block layer and the process scheduler promising better reactivity,” he adds.