Mirrors_BOOKS/LDD-LinuxDeviceDrivers
1
0
Fork 0
mirror of https://github.com/gatieme/LDD-LinuxDeviceDrivers.git synced 2026-08-21 06:24:16 +08:00
Code Issues Actions Packages Projects Releases Wiki Activity

description/scheduler: floating imbalance between NUMA

Browse Source
This commit is contained in:
Cheng Jian
2021-12-13 19:00:08 +08:00
parent a8c8097449
commit 45812c0716
1 changed files with 3 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files
study/kernel/00-DESCRIPTION/SCHEDULER.md
+3 -3
View File
@@ -795,12 +795,12 @@ CPU 负载均衡器在不同的域之间进行平衡, 以分散负载, 并努力
| 时间 | 作者 | 特性 | 描述 | 是否合入主线 | 链接 |
|:----:|:----:|:---:|:---:|:----------:|:----:|
| 2021/09/10 | Kefeng Wang <wangkefeng.wang@huawei.com> | [sched/fair: Allow a small load imbalance between low utilisation SD_NUMA domains](https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=b396f52326de20ec974471b7b19168867b365cbf) | 此补丁允许 NUMA 域之间存在固定程度的两个任务不平衡, 而不管利用率如何. 在许多情况下, 这可以防止通信任务被分离. 一般来说, 补丁只是仅当目标节点几乎完全空闲时才允许出现不平衡, 倾向于在不平衡很小的基本情况下, 寻求避免不必要的跨节点迁移. 对于低利用率的通信工作负载, 这个补丁通常表现得更好, 更少的 NUMA 平衡活动. 对于高使用率而言, 行为没有变化. 评估了不平衡是否应按域大小进行缩放. | v4 ☑ 5.6-rc2 | [PatchWork v4](https://lore.kernel.org/all/20200114101319.GO3466@techsingularity.net), [LKML](https://lkml.org/lkml/2020/1/14/376), [COMMIT](https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=b396f52326de20ec974471b7b19168867b365cbf) |
| 2019/10/18 | Mel Gorman <mgorman@techsingularity.net> | [Reconcile NUMA balancing decisions with the load balancer](https://linuxplumbersconf.org/event/4/contributions/480) | NUMA Balancing 和 Load Blance 经常会做出一些相互冲突的决策(任务放置和页面迁移), 导致不必要的迁移, 这组补丁减少了冲突. 其中 [sched/numa: Use similar logic to the load balancer for moving between domains with spare capacity](https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=fb86f5b2119245afd339280099b4e9417cc0b03a) 引入了 adjust_numa_imbalance() 单独处理和计算 NUMA 层次的不均衡. | v4 ☑ | [LWN](https://lwn.net/Articles/793427), [PatchWork](https://lore.kernel.org/patchwork/cover/1199507), [lkml](https://lkml.org/lkml/2019/10/18/676) |
| 2020/01/14 | Kefeng Wang <wangkefeng.wang@huawei.com> | [sched/fair: Allow a small load imbalance between low utilisation SD_NUMA domains](https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=b396f52326de20ec974471b7b19168867b365cbf) | 此补丁允许 NUMA 域之间存在固定程度的两个任务不平衡, 而不管利用率如何. 在许多情况下, 这可以防止通信任务被分离. 一般来说, 补丁只是仅当目标节点几乎完全空闲时才允许出现不平衡, 倾向于在不平衡很小的基本情况下, 寻求避免不必要的跨节点迁移. 对于低利用率的通信工作负载, 这个补丁通常表现得更好, 更少的 NUMA 平衡活动. 对于高使用率而言, 行为没有变化. 评估了不平衡是否应按域大小进行缩放. | v4 ☑ 5.6-rc2 | [LORE v2](https://lore.kernel.org/lkml/20191220084252.GL3178@techsingularity.net)<br>*-*-*-*-*-*-*-* <br>[PatchWork v4](https://lore.kernel.org/all/20200114101319.GO3466@techsingularity.net), [LKML](https://lkml.org/lkml/2020/1/14/376), [COMMIT](https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=b396f52326de20ec974471b7b19168867b365cbf) |
| 2020/02/24 | Mel Gorman <mgorman@techsingularity.net> | [Reconcile NUMA balancing decisions with the load balancer](https://linuxplumbersconf.org/event/4/contributions/480) | NUMA Balancing 和 Load Blance 经常会做出一些相互冲突的决策(任务放置和页面迁移), 导致不必要的迁移, 这组补丁减少了冲突. 其中 [sched/numa: Use similar logic to the load balancer for moving between domains with spare capacity](https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=fb86f5b2119245afd339280099b4e9417cc0b03a) 引入了 adjust_numa_imbalance() 单独处理和计算 NUMA 层次的不均衡. | v6 ☑ 5.7-rc1 | [LWN](https://lwn.net/Articles/793427), [PatchWork](https://lore.kernel.org/patchwork/cover/1199507) |
| 2020/2/21 | Vincent Guittot | [remove runnable_load_avg and improve group_classify](https://lore.kernel.org/patchwork/cover/1198654) | 重构组调度的 PELT 跟踪, 在每次更新平均负载的时候, 更新整个 CFS_RQ 的平均负载| V10 ☑ 5.7 | [PatchWork](https://lore.kernel.org/patchwork/cover/1198654), [lkml](https://lkml.org/lkml/2020/2/21/1386) |
| 2020/02/03 | Mel Gorman <mgorman@techsingularity.net> | [Accumulated fixes for Load/NUMA Balancing reconcilation series](https://lore.kernel.org/patchwork/cover/1203922) | 解决一个负载平衡问题 | | [PatchWork](https://lore.kernel.org/patchwork/cover/1203922) |
| 2020/02/03 | Mel Gorman <mgorman@techsingularity.net> | [Revisit NUMA imbalance tolerance and fork balancing](https://lore.kernel.org/patchwork/patch/1340184) | 解决一个负载平衡问题<br>之前解决 NUMA Balancing 和 Load Balancing 被调和时, 尝试允许在一定程度的不平衡, 但是也引入了诸多问题. 因此当时限制不平衡只允许在几乎空闲的 NUMA 域. 现在大多数问题都已经解决掉了, 现在[允许不平衡扩大一定的范围](https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=7d2b5dd0bcc48095651f1b85f751eef610b3e034). 同时该补丁还解决了 fork 时候 balance 的问题. 性能测试发现, 这个补丁可以提升约 1.5% unixbench 的跑分, 参见 [e7f28850ea: unixbench.score 1.5% improvement](https://lore.kernel.org/lkml/20201122150415.GJ2390@xsang-OptiPlex-9020) | RFC ☑ 5.11-rc1 | [PatchWork](https://lore.kernel.org/patchwork/cover/1340184) |
| 2021/12/01 | Mel Gorman <mgorman@techsingularity.net> | [Adjust NUMA imbalance for multiple LLCs](https://lore.kernel.org/lkml/20211201151844.20488-1-mgorman@techsingularity.net) | NA | RFC ☐ | [PatchWork v3,0/2](https://lore.kernel.org/lkml/20211201151844.20488-1-mgorman@techsingularity.net) |
| 2021/12/01 | Mel Gorman <mgorman@techsingularity.net> | [Adjust NUMA imbalance for multiple LLCs](https://lore.kernel.org/lkml/20211201151844.20488-1-mgorman@techsingularity.net) | [commit 7d2b5dd0bcc4 ("sched/numa: Allow a floating imbalance between NUMA nodes")](https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=7d2b5dd0bcc4) 允许 NUMA 节点之间的不平衡, 这样通信任务不会被 load balance 分开. 当 LLC 和 node 之间有 1:1 的关系时, 这种方法可以很好地工作, 但是对于多个 LLC, 如果独立的任务过早地使用 CPU 共享缓存, 这种方法就不太理想了. 本系列解决了两个问题:<br>1. 调度程序域权重的使用不一致, 以及当每个 NUMA 节点有许多 LLC 时性能不佳. NUMA之间允许的不均衡的进程数目不再是一个固定的值 NUMA_IMBALANCE_MIN(2), 而是在 build_sched_domains() 中实际探测 NUMA 域下辖的 LLC 的数目, 作为 sd->imb_numa_nr. | v4 ☐ | [PatchWork v3,0/2](https://lore.kernel.org/lkml/20211201151844.20488-1-mgorman@techsingularity.net)<br>*-*-*-*-*-*-*-* <br>[LORE v4,0/2](https://lore.kernel.org/lkml/20211210093307.31701-1-mgorman@techsingularity.net) |
此外 load_balance 上还有一些接口层次的变化