Linus Torvalds releases Linux 5.16 rc1 with new performance-enhancing memory tech
(2021/11/15)
- Reference: 1636937323
- News link: https://www.theregister.co.uk/2021/11/15/linux_5_16_rc1/
- Source link:
Linus Torvalds has loosed the first release candidate for version 5.16 of the Linux kernel.
In his [1]Sunday afternoon State of the Kernel announcement , Torvalds indicated the merge window for the new release did not include any "uhhuh, things aren't working and now I need to bisect where they broke" moments. Work therefore proceeded swiftly, so much so that not even travelling for a few days and using a laptop (which Torvalds wrote is "usually fairly painful") delayed progress.
Torvalds wrote that Linux 5.16 will not be a "huge release" but does include a significant new feature. "Memory Folios" is a memory management system that offers "a more efficient and type-safe way to specify 'head of a group of pages', rather than the page pointers and 'compound_head()' and friends".
[2]
As [3]explained by Matthew Wilcox, a long-time contributor to the Linux kernel (and Oracle employee) who focuses on memory, Memory Folios aims "to allow filesystems and the page cache to manage memory in larger chunks than PAGE_SIZE".
[4]
[5]
Doing so has produced some nice numbers. "The multi-page folios offer some improvement to some workloads," Wilcox wrote. "Real workloads (eg building the kernel, running postgres in a steady state, etc.) seem to benefit between 0–10%."
A single addition to the Linux kernel that could improve performance by ten per cent? Where do we sign?
[6]-Werror pain persists as Linus Torvalds issues Linux 5.15rc2
[7]Linus Torvalds admits to 'self-inflicted damage' with -Werror as Linux 5.15 rc1 debuts
[8]When you finish celebrating Linux turning 30, try new Linux 5.14, says Linus Torvalds
Torvalds rated Memory Folios as "unusually core, but they certainly aren't the bulk of the changes".
Those come from the usual suspects – what Torvalds describes as "drivers (gpu, networking, sound and staging stand out, but it's all over) and architecture code".
Real workloads seem to benefit between 0–10%
"Hardware support is the bulk of the code, it gets the bulk of the changes. But we obviously have all the normal other updates, with filesystem, networking, and core kernel code. With documentation and tooling support filling the gaps."
"Anyway, the merge window may have gone about as smoothly as I could hope for," Torvalds wrote, before adding "but let's get the whole stabilization phase started with some serious testing, shall we?"
[9]
You heard the man. ®
Get our [10]Tech Resources
[1] http://lkml.iu.edu/hypermail/linux/kernel/2111.1/06814.html
[2] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/oses&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=2&c=2YZHpis8Ka0kcGiILAwVOQQAAABY&t=ct%3Dns%26unitnum%3D2%26raptor%3Dcondor%26pos%3Dtop%26test%3D0
[3] https://lkml.org/lkml/2021/10/30/263
[4] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/oses&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44YZHpis8Ka0kcGiILAwVOQQAAABY&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0
[5] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/oses&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=3&c=33YZHpis8Ka0kcGiILAwVOQQAAABY&t=ct%3Dns%26unitnum%3D3%26raptor%3Deagle%26pos%3Dmid%26test%3D0
[6] https://www.theregister.com/2021/09/20/linux_5_15_rc2/
[7] https://www.theregister.com/2021/09/13/linux_5_15_rc1/
[8] https://www.theregister.com/2021/08/30/linux_5_14/
[9] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/oses&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44YZHpis8Ka0kcGiILAwVOQQAAABY&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0
[10] https://whitepapers.theregister.com/
In his [1]Sunday afternoon State of the Kernel announcement , Torvalds indicated the merge window for the new release did not include any "uhhuh, things aren't working and now I need to bisect where they broke" moments. Work therefore proceeded swiftly, so much so that not even travelling for a few days and using a laptop (which Torvalds wrote is "usually fairly painful") delayed progress.
Torvalds wrote that Linux 5.16 will not be a "huge release" but does include a significant new feature. "Memory Folios" is a memory management system that offers "a more efficient and type-safe way to specify 'head of a group of pages', rather than the page pointers and 'compound_head()' and friends".
[2]
As [3]explained by Matthew Wilcox, a long-time contributor to the Linux kernel (and Oracle employee) who focuses on memory, Memory Folios aims "to allow filesystems and the page cache to manage memory in larger chunks than PAGE_SIZE".
[4]
[5]
Doing so has produced some nice numbers. "The multi-page folios offer some improvement to some workloads," Wilcox wrote. "Real workloads (eg building the kernel, running postgres in a steady state, etc.) seem to benefit between 0–10%."
A single addition to the Linux kernel that could improve performance by ten per cent? Where do we sign?
[6]-Werror pain persists as Linus Torvalds issues Linux 5.15rc2
[7]Linus Torvalds admits to 'self-inflicted damage' with -Werror as Linux 5.15 rc1 debuts
[8]When you finish celebrating Linux turning 30, try new Linux 5.14, says Linus Torvalds
Torvalds rated Memory Folios as "unusually core, but they certainly aren't the bulk of the changes".
Those come from the usual suspects – what Torvalds describes as "drivers (gpu, networking, sound and staging stand out, but it's all over) and architecture code".
Real workloads seem to benefit between 0–10%
"Hardware support is the bulk of the code, it gets the bulk of the changes. But we obviously have all the normal other updates, with filesystem, networking, and core kernel code. With documentation and tooling support filling the gaps."
"Anyway, the merge window may have gone about as smoothly as I could hope for," Torvalds wrote, before adding "but let's get the whole stabilization phase started with some serious testing, shall we?"
[9]
You heard the man. ®
Get our [10]Tech Resources
[1] http://lkml.iu.edu/hypermail/linux/kernel/2111.1/06814.html
[2] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/oses&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=2&c=2YZHpis8Ka0kcGiILAwVOQQAAABY&t=ct%3Dns%26unitnum%3D2%26raptor%3Dcondor%26pos%3Dtop%26test%3D0
[3] https://lkml.org/lkml/2021/10/30/263
[4] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/oses&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44YZHpis8Ka0kcGiILAwVOQQAAABY&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0
[5] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/oses&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=3&c=33YZHpis8Ka0kcGiILAwVOQQAAABY&t=ct%3Dns%26unitnum%3D3%26raptor%3Deagle%26pos%3Dmid%26test%3D0
[6] https://www.theregister.com/2021/09/20/linux_5_15_rc2/
[7] https://www.theregister.com/2021/09/13/linux_5_15_rc1/
[8] https://www.theregister.com/2021/08/30/linux_5_14/
[9] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/oses&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44YZHpis8Ka0kcGiILAwVOQQAAABY&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0
[10] https://whitepapers.theregister.com/
Does this mean hugepages are obsolete? Please for the love of god let this mean hugepages are obsolete. And the kludge-upon-kludge that is transparent huge pages. If there's a place linux fucked up architectural design, it was with hugepages.
Seriously, tell anyone "You want 2MB pages as well as 4K? Well, all you have to do is specify the number you want during kernel boot (which is static and must be guessed ahead of time and cannot be changed while the kernel is running), which removes them from the standard memory pool, and allows access by mounting a special file system in /dev/hugepages. Oh, and obviously just using malloc() in 2MB chunks won't work."
You should be spat on.