SoylentNews

Arthur T Knackerbracket [soylentnews.org] writes:

--- --- --- --- Entire Story Below - Must Be Edited --- --- --- --- --- --- ---

Arthur T Knackerbracket has processed the following story [theregister.com]:

Over the weekend, the world's most famous Finn pushed out the latest version of the Linux kernel – and warned of upcoming disruption.

Linux kernel 6.16 was released [kernel.org] after what was apparently a relaxed end of the development cycle. (We suppose this could be interpreted as a subtle dig at certain file system developers [theregister.com], but then again, Torvalds is not famed for subtlety [theregister.com].)

As kernel releases go, this one is almost unusually modest. It doesn't have any huge blockbuster new features, but does contain a large number of bugfixes and code. Phoronix estimates [phoronix.com] that it has 38.4 million lines of code across over 78,000 files. Remember when the central design ethos of UNIX was that it was small and simple and clean? Well, no, me neither, because around the time The Reg FOSS desk first touched a computer keyboard, UNIX System III [gunkies.org] came out, one of the first releases that unified different codebases, and also one of the first commercial editions from AT&T. But that was the idea, right?

Kernel 6.16 supports Intel's 2023 Advanced Performance Extensions [theregister.com], which means improved vector instructions and doubling the number of general-purpose registers available. (Only certain CPU models benefit from the full-width version of the new vector instructions, though, which is arguably an example of the sort of moves that caused Intel to falter in recent years.)

Two of the built-in file systems get performance tweaks that allow for larger individual blocks of data. XFS, open sourced by SGI at the turn of the century [theregister.com], now gets larger atomic writes [lwn.net]. Meanwhile, ext4 gets bigalloc [kernel.org] and large folio support [kernel.org], which can make some operations about one-third faster. Btrfs and NFS both get tweaks, too.

On pretty much any Unix, when a program crashes, it emits a core dump [archlinux.org] and saves it in the current working directory. Among other improvements [lwn.net], now a core can be sent over an AF_SOCKET [man7.org] instead. This means both functional improvements as well as security ones.

On big iron, Linux's support for NUMA systems, which The Register explained when AMD brought it to x86 [theregister.com], now can automatically self-tune [lwn.net], among other optimizations. Support for five-level page tables allows for enormous amounts of virtual memory, as LWN's 2017 article explains [lwn.net].

On small iron, the kernel can now offload sound decoding to USB hardware, catching up [lwn.net] with onboard sound chips – a change that's taken years [phoronix.com] to make it in.

We can't help but feel that these two demonstrate the remarkable range of kit that Linux is used for. No wonder it's got so big, really.

The sound offload explanation we linked to above came from Linux Weekly News's two-part round-up of what was new for this merge window: the first half [lwn.net] and the second half [lwn.net] are linked, for the real nitty-gritty, as well as an overview [lwn.net]. The "kernel newbies" site has one big summary [kernelnewbies.org] for the truly hardcore.

Linus also noted in his announcement that he will be traveling for a lot of the 6.17 merge window, which might cause disruption. On the one hand, this might count as a warning to developers, but it's also a reminder that there's one man at the very top of this pyramid of developers. It leads us to wonder if the next big change in OS kernel development will come when he retires, rather than any technological milestone. ®

Original Submission