Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]
Groups > linux.debian.kernel > #51303 > unrolled thread
| Started by | Quentin Deldycke <quentindeldycke@gmail.com> |
|---|---|
| First post | 2015-11-12 14:20 +0100 |
| Last post | 2015-11-12 18:40 +0100 |
| Articles | 2 — 2 participants |
Back to article view | Back to linux.debian.kernel
Bug#804857: linux: New feature: enable CONFIG_NO_HZ_FULL and CONFIG_RCU_NOCB_CPU/CONFIG_RCU_NOCB_CPU_NONE Quentin Deldycke <quentindeldycke@gmail.com> - 2015-11-12 14:20 +0100
Bug#804857: linux: New feature: enable CONFIG_NO_HZ_FULL and CONFIG_RCU_NOCB_CPU/CONFIG_RCU_NOCB_CPU_NONE Ben Hutchings <ben@decadent.org.uk> - 2015-11-12 18:40 +0100
| From | Quentin Deldycke <quentindeldycke@gmail.com> |
|---|---|
| Date | 2015-11-12 14:20 +0100 |
| Subject | Bug#804857: linux: New feature: enable CONFIG_NO_HZ_FULL and CONFIG_RCU_NOCB_CPU/CONFIG_RCU_NOCB_CPU_NONE |
| Message-ID | <qtZXc-VN-13@gated-at.bofh.it> |
Source: linux Severity: wishlist Dear Maintainer, Could you please enable kernel options CONFIG_NO_HZ_FULL, CONFIG_RCU_NOCB_CPU and CONFIG_RCU_NOCB_CPU_NONE. As explained in: https://www.kernel.org/doc/Documentation/timers/NO_HZ.txt This options: - Don't modify default kernel behavior - Enable kernel command line nohz_full to enable tickless cpus - Enable kernel command line rcu_nocbs to offload rcu callbacks The combination of these parameters + cpu-pinning of virtual machines greatly improve performance. It also reduces latency in these virtual machines as the host system doesn't try to use it. (if your cgroup is correctly made) These parameters are disabled in all debian kernels. These parameters are enabled for various linux distribution by default without problem. (rhel 7 being one of them) (note that rhel 7 uses CONFIG_RCU_NOCB_CPU_ALL and user can pin offload kthreads to a core) The usage of CONFIG_RCU_NOCB_CPU_NONE doesn't offload by default. But possibly this behavior is not the best. Quentin. -- System Information: Debian Release: stretch/sid APT prefers stable-updates APT policy: (500, 'stable-updates'), (500, 'unstable'), (500, 'testing'), (500, 'stable'), (500, 'oldstable'), (1, 'experimental') Architecture: amd64 (x86_64) Foreign Architectures: i386 Kernel: Linux 4.3.0-rc7-amd64 (SMP w/8 CPU cores) Locale: LANG=en_US.UTF-8, LC_CTYPE=en_US.UTF-8 (charmap=UTF-8) Shell: /bin/sh linked to /bin/dash Init: systemd (via /run/systemd/system)
[toc] | [next] | [standalone]
| From | Ben Hutchings <ben@decadent.org.uk> |
|---|---|
| Date | 2015-11-12 18:40 +0100 |
| Message-ID | <qu40O-3rz-19@gated-at.bofh.it> |
| In reply to | #51303 |
[Multipart message — attachments visible in raw view] — view raw
On Thu, 2015-11-12 at 14:13 +0100, Quentin Deldycke wrote: > Source: linux > Severity: wishlist > > Dear Maintainer, > > Could you please enable kernel options CONFIG_NO_HZ_FULL, > CONFIG_RCU_NOCB_CPU and CONFIG_RCU_NOCB_CPU_NONE. > > As explained in: https://www.kernel.org/doc/Documentation/timers/NO_HZ.txt > > This options: > - Don't modify default kernel behavior > - Enable kernel command line nohz_full to enable tickless cpus > - Enable kernel command line rcu_nocbs to offload rcu callbacks > > The combination of these parameters + cpu-pinning of virtual machines > greatly improve performance. > It also reduces latency in these virtual machines as the host system > doesn't try to use it. (if your cgroup is correctly made) [...] But for other configurations it is worse: config NO_HZ_FULL bool "Full dynticks system (tickless)" ... This is implemented at the expense of some overhead in user <-> kernel transitions: syscalls, exceptions and interrupts. Even when it's dynamically off. Say N. I think we should enable this for the rt featureset when it's next available, but not by default. Ben. -- Ben Hutchings Logic doesn't apply to the real world. - Marvin Minsky
[toc] | [prev] | [standalone]
Back to top | Article view | linux.debian.kernel
csiph-web