mirror of
https://github.com/gatieme/LDD-LinuxDeviceDrivers.git
synced 2026-08-21 06:24:16 +08:00
description/scheduler: NO_HZ
This commit is contained in:
Cheng Jian
@@ -62,6 +62,7 @@ blogexcerpt: 虚拟化 & KVM 子系统
|
||||
| 时间 | 作者 | 特性 | 描述 | 是否合入主线 | 链接 |
|
||||
|:----:|:----:|:---:|:----:|:---------:|:----:|
|
||||
| 2021/06/01 | Shaokun Zhang <zhangshaokun@hisilicon.com> | [fs: Optimized file struct to improve performance](https://patchwork.kernel.org/project/linux-fsdevel/patch%20/1622513557-46189-1-git-send-email-zhangshaokun@hisilicon.com) | 通过调整 struct file 中各字段的布局来提升性能. 在系统调用过程中, 经常使用 struct file 结构体中 `f_count`和 `f_mod` 两个[字段](https://patchwork.kernel.org/project/linux-fsdevel/patch/1592987548-8653-1-git-send-email-zhangshaokun@hisilicon.com), 如果我们将它们放在一起, 将能有效地共享同一 cache line, 这对性能非常有用. intel 0-day CI 工程发现该补丁可以提升 UnixBench System Call Overhead 子项的性能, 参见 [aec499039e: unixbench.score 19.2% improvement](https://lkml.org/lkml/2021/4/20/28). | v1 ☐ | [2021/04/09 Patchwork RESEND](https://patchwork.kernel.org/project/linux-fsdevel/patch/1617940057-52843-1-git-send-email-zhangshaokun@hisilicon.com)<br>*-*-*-*-*-*-*-* <br>[2021/06/01 Patchwork RESEND](https://patchwork.kernel.org/project/linux-fsdevel/patch%20/1622513557-46189-1-git-send-email-zhangshaokun@hisilicon.com) |
|
||||
| 2021/08/03 | Peter Oskolkov <posk@google.com> | [5-10% increase in IO latencies with nohz balance patch](https://lkml.org/lkml/2021/11/29/1108) | [7fd7a9e0caba ("sched/fair: Trigger nohz.next_balance updates when a CPU goes NOHZ-idle")](https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=7fd7a9e0caba) 这个补丁导致了 FIO 性能测试劣化. | v1 ☐ | [PatchWork v4,00/10](https://lore.kernel.org/patchwork/cover/1471548) |
|
||||
|
||||
|
||||
# 2 网络
|
||||
|
||||
@@ -542,7 +542,8 @@ https://lore.kernel.org/lkml/157476581065.5793.4518979877345136813.stgit@buzz/
|
||||
| 时间 | 作者 | 特性 | 描述 | 是否合入主线 | 链接 |
|
||||
|:----:|:----:|:---:|:----:|:---------:|:----:|
|
||||
| 2021/06/21 | JHuaixin Chang | [sched/fair: Burstable CFS bandwidth controller](https://lore.kernel.org/patchwork/cover/1396878) | 突发任务的带宽控制优化, 通过临时之前剩余累计的配额, 使得突发进程在当前周期的配额突然用尽之后, 还可以临时使用之前累计的配额使用, 从而降低突发任务的时延. | v6 ☑ 5.14-rc1 | [ 2020/12/17 v1](https://lore.kernel.org/patchwork/cover/1354613)<br>*-*-*-*-*-*-*-*<br>[2021/01/20 v2](https://lore.kernel.org/patchwork/cover/1368037)<br>*-*-*-*-*-*-*-*<br>[2021/01/21 v3](https://lore.kernel.org/patchwork/cover/1368746)<br>*-*-*-*-*-*-*-*<br>[2021-02-02 v4](https://lore.kernel.org/patchwork/cover/1396878)<br>*-*-*-*-*-*-*-*<br>[2021/05/20 v5](https://lore.kernel.org/patchwork/cover/1433660)<br>*-*-*-*-*-*-*-*<br>[2021/06/21 v6](https://lore.kernel.org/patchwork/cover/1449268)<br>*-*-*-*-*-*-*-*<br>[commit](https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=f4183717b370ad28dd0c0d74760142b20e6e7931) |
|
||||
| 2021/08/30 | Huaixin Chang <changhuaixin@linux.alibaba.com> | [Add statistics and ducument for cfs bandwidth burst](https://lore.kernel.org/patchwork/cover/1396878) | 为 Burstable CFS bandwidth 添加统计信息和文档. | v1 ☐ | [ 2020/12/17 v1](https://lore.kernel.org/patchwork/cover/1396878)<br>*-*-*-*-*-*-*-*<br>[2021/08/30 LORE v2 0/2](https://lore.kernel.org/all/20210830032215.16302-1-changhuaixin@linux.alibaba.com) |
|
||||
| 2021/08/30 | Huaixin Chang <changhuaixin@linux.alibaba.com> | [Add statistics and ducument for cfs bandwidth burst](https://lore.kernel.org/patchwork/cover/1396878) | 为 Burstable CFS bandwidth 添加统计信息和文档. | v1 ☑ 5.16-rc1 | [2020/12/17 v1](https://lore.kernel.org/patchwork/cover/1396878)<br>*-*-*-*-*-*-*-*<br>[2021/08/30 LORE v2 0/2](https://lore.kernel.org/all/20210830032215.16302-1-changhuaixin@linux.alibaba.com) |
|
||||
| 2021/11/29 | Honglei Wang <wanghonglei@didichuxing.com> | [sched/fair: prevent cpu burst too many periods](https://lore.kernel.org/patchwork/cover/1396878) | commit f4183717b370 ("sched/fair: Introduce the burstable CFS controller") 引入了一个问题, 任务在持久性期间可能获得比配额更多的 cpu. 例如, 一个任务组的配额为每周期 100ms, 可以获得 100ms 突发, 其平均利用率约为每周期 105ms. 一旦这个组获得了一个空闲时间段, 它就有机会在公共带宽配置中获得超过其配额的 10 个或更多时间段的计算能力(例如, 100 毫秒作为时间段). 这意味着任务获得了可以 "偷走" 完成日常工作的能力, 因为所有任务都可以安排出去或睡觉, 以帮助团队获得空闲时间. cpu burst 的本来目的是帮助处理突发性工作负载. 但是, 如果一个任务组在没有突发性工作负载的情况下, 能够在持续时间内获得超过其配额的计算能力, 那么它违背了初衷. 此修补程序将突发限制为一个时段, 以便在很长时间内不会突破配额限制. 有了这个, 我们可以给任务组更多的 cpu 突发能力来处理真正的突发性工作负载， 而不必担心被恶意 "窃取". | v1 ☑ 5.16-rc1 | [LKML](https://lkml.org/lkml/2021/11/29/663) |
|
||||
|
||||
|
||||
## 2.2 实时进程的组调度支持(RT Group Scheduling)
|
||||
@@ -1323,6 +1324,14 @@ NOHZ 社区测试用例 [frederic/dynticks-testing.git](https://git.kernel.org/p
|
||||
### 8.2.2 1 HZ tick
|
||||
-------
|
||||
|
||||
之前当 CPU 运行在 full dynticks(NO_HZ_FULL)模式时, 仍旧会保持 1Hz 的 tick, 以保持调度程序统计有效. [sched: Keep at least 1 tick per second for active dynticks tasks](https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=265f22a975c1e4cc3a4d1f94a3ec53ffbb6f5b9f).
|
||||
|
||||
然而, 这些残留的噪音干扰对于不能忍受任何中断或想要最小化背景干扰的应用来说是依旧是一种负担.
|
||||
|
||||
[sched/fair: Update blocked load from NEWIDLE](https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=e022e0d38ad475fc650f22efa3deb2fb96e62542), 允许其他 CPU 远程更新 NEWIDLE CPU 的 blocked average.
|
||||
|
||||
接着 [sched/isolation: Offload residual 1Hz scheduler tick](https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=d84b31313ef8a8de55a2cbfb72f76f36d8c927fb) 允许调度器将 TICK 外包给全局工作队列, 以便其他 CPU 可以远程处理这些 TICK. 至此, kernel 支持了允许将原本的 1HZ tick 移除掉.
|
||||
|
||||
| 时间 | 作者 | 特性 | 描述 | 是否合入主线 | 链接 |
|
||||
|:----:|:----:|:---:|:---:|:----------:|:----:|
|
||||
| 2015/06/11 | Frederic Weisbecker <fweisbec@gmail.com> | [nohz: Bunch of fixes](https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=594493594373) | NA | v1 ☑ 4.3-rc1 | [LKML 0/2](https://lkml.org/lkml/2013/5/3/374) |
|
||||
|
||||
Reference in New Issue
Block a user