你不好奇 CPU 是如何执行任务的吗？

{"type":"doc","content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https://static001.geekbang.org/infoq/a6/a6fee5ac5c97ecd3aceb7db6bb88c423.png","alt":null,"title":null,"style":null,"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"horizontalrule"},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"heading","attrs":{"align":null,"level":2},"content":[{"type":"text","text":"前言"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"你清楚下面这几个问题吗？"}]},{"type":"bulletedlist","content":[{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"有了内存，为什么还需要 CPU Cache？"}]}]},{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"CPU 是怎么读写数据的？"}]}]},{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"如何让 CPU 能读取数据更快一些？"}]}]},{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"CPU 伪共享是如何发生的？又该如何避免？"}]}]},{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"CPU 是如何调度任务的？如果你的任务对响应要求很高，你希望它总是能被先调度，这该怎么办？"}]}]},{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"..."}]}]}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"这篇，我们就来回答这些问题。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https://static001.geekbang.org/infoq/bb/bbc00808f7fb4a7772983153793115f5.png","alt":null,"title":null,"style":null,"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"horizontalrule"},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"heading","attrs":{"align":null,"level":2},"content":[{"type":"text","text":"正文"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"heading","attrs":{"align":null,"level":3},"content":[{"type":"text","text":"CPU 如何读写数据的？"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"先来认识 CPU 的架构，只有理解了 CPU 的 架构，才能更好地理解 CPU 是如何读写数据的，对于现代 CPU 的架构图如下："}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https://static001.geekbang.org/infoq/c8/c82e77a1cfb8ce58dc5ccd5887c14b8c.png","alt":null,"title":null,"style":null,"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"可以看到，一个 CPU 里通常会有多个 CPU 核心，比如上图中的 1 号和 2 号 CPU 核心，并且每个 CPU 核心都有自己的 L1 Cache 和 L2 Cache，而 L1 Cache 通常分为 dCache（数据缓存） 和 iCache（指令缓存），L3 Cache 则是多个核心共享的，这就是 CPU 典型的缓存层次。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"上面提到的都是 CPU 内部的 Cache，放眼外部的话，还会有内存和硬盘，这些存储设备共同构成了金字塔存储层次。如下图所示："}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https://static001.geekbang.org/infoq/31/3195a8229fabb315f894ee6751d10121.png","alt":null,"title":null,"style":null,"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"从上图也可以看到，从上往下，存储设备的容量会越大，而访问速度会越慢。至于每个存储设备的访问延时，你可以看下图的表格："}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https://static001.geekbang.org/infoq/71/7173130778f95493ece4bb181a4f1684.png","alt":null,"title":null,"style":null,"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"你可以看到， CPU 访问 L1 Cache 速度比访问内存快 100 倍，这就是为什么 CPU 里会有 L1~L3 Cache 的原因，目的就是把 Cache 作为 CPU 与内存之间的缓存层，以减少对内存的访问频率。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"CPU 从内存中读取数据到 Cache 的时候，并不是一个字节一个字节读取，而是一块一块的方式来读取数据的，这一块一块的数据被称为 CPU Line（缓存行），所以 "},{"type":"text","marks":[{"type":"strong"}],"text":"CPU Line 是 CPU 从内存读取数据到 Cache 的单位"},{"type":"text","text":"。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"至于 CPU Line 大小，在 Linux 系统可以用下面的方式查看到，你可以看我服务器的 L1 Cache Line 大小是 64 字节，也就意味着 "},{"type":"text","marks":[{"type":"strong"}],"text":"L1 Cache 一次载入数据的大小是 64 字节"},{"type":"text","text":"。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https://static001.geekbang.org/infoq/da/daf796cdc52796bed78d34c738d51ef9.png","alt":null,"title":null,"style":null,"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":" 那么对数组的加载， CPU 就会加载数组里面连续的多个数据到 Cache 里，因此我们应该按照物理内存地址分布的顺序去访问元素，这样访问数组元素的时候，Cache 命中率就会很高，于是就能减少从内存读取数据的频率， 从而可提高程序的性能。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":" "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":" 但是，在我们不使用数组，而是使用单独的变量的时候，则会有 Cache 伪共享的问题，Cache 伪共享问题上是一个性能杀手，我们应该要规避它。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"接下来，就来看看 Cache 伪共享是什么？又如何避免这个问题？"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"现在假设有一个双核心的 CPU，这两个 CPU 核心并行运行着两个不同的线程，它们同时从内存中读取两个不同的数据，分别是类型为 "},{"type":"codeinline","content":[{"type":"text","text":"long"}]},{"type":"text","text":" 的变量 A 和 B，这个两个数据的地址在物理内存上是"},{"type":"text","marks":[{"type":"strong"}],"text":"连续"},{"type":"text","text":"的，如果 Cahce Line 的大小是 64 字节，并且变量 A 在 Cahce Line 的开头位置，那么这两个数据是位于"},{"type":"text","marks":[{"type":"strong"}],"text":"同一个 Cache Line 中"},{"type":"text","text":"，又因为 CPU Line 是 CPU 从内存读取数据到 Cache 的单位，所以这两个数据会被同时读入到了两个 CPU 核心中各自 Cache 中。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https://static001.geekbang.org/infoq/3c/3cb049f3623c4172365d750e6a25cf5c.png","alt":null,"title":null,"style":null,"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"我们来思考一个问题，如果这两个不同核心的线程分别修改不同的数据，比如 1 号 CPU 核心的线程只修改了 变量 A，或 2 号 CPU 核心的线程的线程只修改了变量 B，会发生什么呢？"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"heading","attrs":{"align":null,"level":4},"content":[{"type":"text","text":"分析伪共享的问题"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"现在我们结合保证多核缓存一致的 MESI 协议，来说明这一整个的过程，如果你还不知道 MESI 协议，你可以看我这篇文章「"},{"type":"link","attrs":{"href":"https://mp.weixin.qq.com/s/PDUqwAIaUxNkbjvRfovaCg","title":""},"content":[{"type":"text","text":"10 张图打开 CPU 缓存一致性的大门"}]},{"type":"text","text":"」。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"①. 最开始变量 A 和 B 都还不在 Cache 里面，假设 1 号核心绑定了线程 A，2 号核心绑定了线程 B，线程 A 只会读写变量 A，线程 B 只会读写变量 B。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https://static001.geekbang.org/infoq/51/51b2301b0b10b574dea4dc532eee6dc3.png","alt":null,"title":null,"style":null,"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"②. 1 号核心读取变量 A，由于 CPU 从内存读取数据到 Cache 的单位是 Cache Line，也正好变量 A 和 变量 B 的数据归属于同一个 Cache Line，所以 A 和 B 的数据都会被加载到 Cache，并将此 Cache Line 标记为「独占」状态。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https://static001.geekbang.org/infoq/57/572fcfaf7c029b1957b7ae9bf07397ad.png","alt":null,"title":null,"style":null,"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"③. 接着，2 号核心开始从内存里读取变量 B，同样的也是读取 Cache Line 大小的数据到 Cache 中，此 Cache Line 中的数据也包含了变量 A 和 变量 B，此时 1 号和 2 号核心的 Cache Line 状态变为「共享」状态。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https://static001.geekbang.org/infoq/c8/c80c098b80caa70142fecf1848672f22.png","alt":null,"title":null,"style":null,"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"④. 1 号核心需要修改变量 A，发现此 Cache Line 的状态是「共享」状态，所以先需要通过总线发送消息给 2 号核心，通知 2 号核心把 Cache 中对应的 Cache Line 标记为「已失效」状态，然后 1 号核心对应的 Cache Line 状态变成「已修改」状态，并且修改变量 A。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https://static001.geekbang.org/infoq/36/369dd0c954e7d87029ac6a5332e7711c.png","alt":null,"title":null,"style":null,"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"⑤. 之后，2 号核心需要修改变量 B，此时 2 号核心的 Cache 中对应的 Cache Line 是已失效状态，另外由于 1 号核心的 Cache 也有此相同的数据，且状态为「已修改」状态，所以要先把 1 号核心的 Cache 对应的 Cache Line 写回到内存，然后 2 号核心再从内存读取 Cache Line 大小的数据到 Cache 中，最后把变量 B 修改到 2 号核心的 Cache 中，并将状态标记为「已修改」状态。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https://static001.geekbang.org/infoq/7a/7a27b4aa80c086e69ffe2afa83601c52.png","alt":null,"title":null,"style":null,"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"所以，可以发现如果 1 号和 2 号 CPU 核心这样持续交替的分别修改变量 A 和 B，就会重复 ④ 和 ⑤ 这两个步骤，Cache 并没有起到缓存的效果，虽然变量 A 和 B 之间其实并没有任何的关系，但是因为同时归属于一个 Cache Line ，这个 Cache Line 中的任意数据被修改后，都会相互影响，从而出现 ④ 和 ⑤ 这两个步骤。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"因此，这种因为多个线程同时读写同一个 Cache Line 的不同变量时，而导致 CPU Cache 失效的现象称为*"},{"type":"text","marks":[{"type":"italic"}],"text":"伪共享（"},{"type":"text","text":"False Sharing"},{"type":"text","marks":[{"type":"italic"}],"text":"）"},{"type":"text","text":"*。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"heading","attrs":{"align":null,"level":4},"content":[{"type":"text","text":"避免伪共享的方法"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"因此，对于多个线程共享的热点数据，即经常会修改的数据，应该避免这些数据刚好在同一个 Cache Line 中，否则就会出现为伪共享的问题。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"接下来，看看在实际项目中是用什么方式来避免伪共享的问题的。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"在 Linux 内核中存在 "},{"type":"codeinline","content":[{"type":"text","text":"__cacheline_aligned_in_smp"}]},{"type":"text","text":" 宏定义，是用于解决伪共享的问题。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https://static001.geekbang.org/infoq/18/18d82a849ded6f81e4dee103a4fa8d94.png","alt":null,"title":null,"style":null,"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"从上面的宏定义，我们可以看到："}]},{"type":"bulletedlist","content":[{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"如果在多核（MP）系统里，该宏定义是 "},{"type":"codeinline","content":[{"type":"text","text":"__cacheline_aligned"}]},{"type":"text","text":"，也就是 Cache Line 的大小；"}]}]},{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"而如果在单核系统里，该宏定义是空的；"}]}]}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"因此，针对在同一个 Cache Line 中的共享的数据，如果在多核之间竞争比较严重，为了防止伪共享现象的发生，可以采用上面的宏定义使得变量在 Cache Line 里是对齐的。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"举个例子，有下面这个结构体："}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https://static001.geekbang.org/infoq/cd/cde18c30d1719c85359be4884fd93a7b.png","alt":null,"title":null,"style":null,"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"结构体里的两个成员变量 a 和 b 在物理内存地址上是连续的，于是它们可能会位于同一个 Cache Line 中，如下图："}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https://static001.geekbang.org/infoq/fc/fcf28152b7fa66399f2b0d97c63c2f53.png","alt":null,"title":null,"style":null,"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"所以，为了防止前面提到的 Cache 伪共享问题，我们可以使用上面介绍的宏定义，将 b 的地址设置为 Cache Line 对齐地址，如下："}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https://static001.geekbang.org/infoq/c6/c6fecac537be5e50574b69c3294d331d.png","alt":null,"title":null,"style":null,"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"这样 a 和 b 变量就不会在同一个 Cache Line 中了，如下图："}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https://static001.geekbang.org/infoq/7e/7e1731340bb1e41f8c12bf5a20e2f3b4.png","alt":null,"title":null,"style":null,"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"所以，避免 Cache 伪共享实际上是用空间换时间的思想，浪费一部分 Cache 空间，从而换来性能的提升。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"我们再来看一个应用层面的规避方案，有一个 Java 并发框架 Disruptor 使用「字节填充 + 继承」的方式，来避免伪共享的问题。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"Disruptor 中有一个 RingBuffer 类会经常被多个线程使用，代码如下："}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https://static001.geekbang.org/infoq/ac/acef1063a14deea9f049f19c8b4e950b.png","alt":null,"title":null,"style":null,"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"你可能会觉得 RingBufferPad 类里 7 个 long 类型的名字很奇怪，但事实上，它们虽然看起来毫无作用，但却对性能的提升起到了至关重要的作用。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"我们都知道，CPU Cache 从内存读取数据的单位是 CPU Line，一般 64 位 CPU 的 CPU Line 的大小是 64 个字节，一个 long 类型的数据是 8 个字节，所以 CPU 一下会加载 8 个 long 类型的数据。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"根据 JVM 对象继承关系中父类成员和子类成员，内存地址是连续排列布局的，因此 RingBufferPad 中的 7 个 long 类型数据作为 Cache Line "},{"type":"text","marks":[{"type":"strong"}],"text":"前置填充"},{"type":"text","text":"，而 RingBuffer 中的 7 个 long 类型数据则作为 Cache Line "},{"type":"text","marks":[{"type":"strong"}],"text":"后置填充"},{"type":"text","text":"，这 14 个 long 变量没有任何实际用途，更不会对它们进行读写操作。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https://static001.geekbang.org/infoq/a0/a00e961a867d40662ea2248a018bc0ff.png","alt":null,"title":null,"style":null,"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"另外，RingBufferFelds 里面定义的这些变量都是 "},{"type":"codeinline","content":[{"type":"text","text":"final"}]},{"type":"text","text":" 修饰的，意味着第一次加载之后不会再修改， 又"},{"type":"text","marks":[{"type":"strong"}],"text":"由于「前后」各填充了 7 个不会被读写的 long 类型变量，所以无论怎么加载 Cache Line，这整个 Cache Line 里都没有会发生更新操作的数据，于是只要数据被频繁地读取访问，就自然没有数据被换出 Cache 的可能，也因此不会产生伪共享的问题"},{"type":"text","text":"。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"horizontalrule"},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"heading","attrs":{"align":null,"level":3},"content":[{"type":"text","text":"CPU 如何选择线程的？"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"了解完 CPU 读取数据的过程后，我们再来看看 CPU 是根据什么来选择当前要执行的线程。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"在 Linux 内核中，进程和线程都是用 "},{"type":"codeinline","content":[{"type":"text","text":"tark_struct"}]},{"type":"text","text":" 结构体表示的，区别在于线程的 tark"},{"type":"text","marks":[{"type":"italic"}],"text":"struct 结构体里部分资源是共享了进程已创建的资源，比如内存地址空间、代码段、文件描述符等，所以 Linux 中的线程也被称为轻量级进程，因为线程的 tark"},{"type":"text","text":"struct 相比进程的 tark_struct 承载的 资源比较少，因此以「轻」得名。 "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"一般来说，没有创建线程的进程，是只有单个执行流，它被称为是主线程。如果想让进程处理更多的事情，可以创建多个线程分别去处理，但不管怎么样，它们对应到内核里都是 "},{"type":"codeinline","content":[{"type":"text","text":"tark_struct"}]},{"type":"text","text":"。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https://static001.geekbang.org/infoq/40/40babd581cbb64638c323e1a6fbdd4d9.png","alt":null,"title":null,"style":null,"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"所以，Linux 内核里的调度器，调度的对象就是 "},{"type":"codeinline","content":[{"type":"text","text":"tark_struct"}]},{"type":"text","text":"，接下来我们就把这个数据结构统称为"},{"type":"text","marks":[{"type":"strong"}],"text":"任务"},{"type":"text","text":"。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"在 Linux 系统中，根据任务的优先级以及响应要求，主要分为两种，其中优先级的数值越小，优先级越高："}]},{"type":"bulletedlist","content":[{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"实时任务，对系统的响应时间要求很高，也就是要尽可能快的执行实时任务，优先级在 "},{"type":"codeinline","content":[{"type":"text","text":"0~99"}]},{"type":"text","text":" 范围内的就算实时任务；"}]}]},{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"普通任务，响应时间没有很高的要求，优先级在 "},{"type":"codeinline","content":[{"type":"text","text":"100~139"}]},{"type":"text","text":" 范围内都是普通任务级别；"}]}]}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"heading","attrs":{"align":null,"level":4},"content":[{"type":"text","text":"调度类"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"由于任务有优先级之分，Linux 系统为了保障高优先级的任务能够尽可能早的被执行，于是分为了这几种调度类，如下图："}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https://static001.geekbang.org/infoq/d6/d62c16c921cd7bbe943a00eb2f34de77.png","alt":null,"title":null,"style":null,"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"Deadline 和 Realtime 这两个调度类，都是应用于实时任务的，这两个调度类的调度策略合起来共有这三种，它们的作用如下： "}]},{"type":"bulletedlist","content":[{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"italic"}],"text":"SCHED_DEADLINE"},{"type":"text","text":"：是按照 deadline 进行调度的，距离当前时间点最近的 deadline 的任务会被优先调度；"}]}]},{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"italic"}],"text":"SCHED_FIFO"},{"type":"text","text":"：对于相同优先级的任务，按先来先服务的原则，但是优先级更高的任务，可以抢占低优先级的任务，也就是优先级高的可以「插队」；"}]}]},{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"italic"}],"text":"SCHED_RR"},{"type":"text","text":"：对于相同优先级的任务，轮流着运行，每个任务都有一定的时间片，当用完时间片的任务会被放到队列尾部，以保证相同优先级任务的公平性，但是高优先级的任务依然可以抢占低优先级的任务；"}]}]}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"而 Fair 调度类是应用于普通任务，都是由 CFS 调度器管理的，分为两种调度策略："}]},{"type":"bulletedlist","content":[{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"italic"}],"text":"SCHED_NORMAL"},{"type":"text","text":"：普通任务使用的调度策略；"}]}]},{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"italic"}],"text":"SCHED_BATCH"},{"type":"text","text":"：后台任务的调度策略，不和终端进行交互，因此在不影响其他需要交互的任务，可以适当降低它的优先级。"}]}]}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"heading","attrs":{"align":null,"level":4},"content":[{"type":"text","text":"完全公平调度"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"我们平日里遇到的基本都是普通任务，对于普通任务来说，公平性最重要，在 Linux 里面，实现了一个基于 CFS 的调度算法，也就是*"},{"type":"text","marks":[{"type":"italic"}],"text":"完全公平调度（"},{"type":"text","text":"Completely Fair Scheduling"},{"type":"text","marks":[{"type":"italic"}],"text":"）"},{"type":"text","text":"*。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"这个算法的理念是想让分配给每个任务的 CPU 时间是一样，于是它为每个任务安排一个虚拟运行时间 vruntime，如果一个任务在运行，其运行的越久，该任务的 vruntime 自然就会越大，而没有被运行的任务，vruntime 是不会变化的。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"那么，"},{"type":"text","marks":[{"type":"strong"}],"text":"在 CFS 算法调度的时候，会优先选择 vruntime 少的任务"},{"type":"text","text":"，以保证每个任务的公平性。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"这就好比，让你把一桶的奶茶平均分到 10 杯奶茶杯里，你看着哪杯奶茶少，就多倒一些；哪个多了，就先不倒，这样经过多轮操作，虽然不能保证每杯奶茶完全一样多，但至少是公平的。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"当然，上面提到的例子没有考虑到优先级的问题，虽然是普通任务，但是普通任务之间还是有优先级区分的，所以在计算虚拟运行时间 vruntime 还要考虑普通任务的"},{"type":"text","marks":[{"type":"strong"}],"text":"权重值"},{"type":"text","text":"，注意权重值并不是优先级的值，内核中会有一个 nice 级别与权重值的转换表，nice 级别越低的权重值就越大，至于 nice 值是什么，我们后面会提到。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"于是就有了以下这个公式："}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https://static001.geekbang.org/infoq/c8/c8d9cb54b02030ca3add0121bfebe7bd.png","alt":null,"title":null,"style":null,"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"你可以不用管 NICE"},{"type":"text","marks":[{"type":"italic"}],"text":"0"},{"type":"text","text":"LOAD 是什么，你就认为它是一个常量，那么在「同样的实际运行时间」里，高权重任务的 vruntime 比低权重任务的 vruntime "},{"type":"text","marks":[{"type":"strong"}],"text":"少"},{"type":"text","text":"，你可能会奇怪为什么是少的？你还记得 CFS 调度吗，它是会优先选择 vruntime 少的任务进行调度，所以高权重的任务就会被优先调度了，于是高权重的获得的实际运行时间自然就多了。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"heading","attrs":{"align":null,"level":4},"content":[{"type":"text","text":"CPU 运行队列"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"一个系统通常都会运行着很多任务，多任务的数量基本都是远超 CPU 核心数量，因此这时候就需要"},{"type":"text","marks":[{"type":"strong"}],"text":"排队"},{"type":"text","text":"。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"事实上，每个 CPU 都有自己的*"},{"type":"text","marks":[{"type":"italic"}],"text":"运行队列（"},{"type":"text","text":"Run Queue, rq"},{"type":"text","marks":[{"type":"italic"}],"text":"）"},{"type":"text","text":"*，用于描述在此 CPU 上所运行的所有进程，其队列包含三个运行队列，Deadline 运行队列 dl"},{"type":"text","marks":[{"type":"italic"}],"text":"rq、实时任务运行队列 rt"},{"type":"text","text":"rq 和 CFS 运行队列 csf"},{"type":"text","marks":[{"type":"italic"}],"text":"rq，其中 csf"},{"type":"text","text":"rq 是用红黑树来描述的，按 vruntime 大小来排序的，最左侧的叶子节点，就是下次会被调度的任务。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https://static001.geekbang.org/infoq/d8/d8470503a7e83a9dee3544dc0a0dc13b.png","alt":null,"title":null,"style":null,"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"这几种调度类是有优先级的，优先级如下：Deadline > Realtime > Fair，这意味着 Linux 选择下一个任务执行的时候，会按照此优先级顺序进行选择，也就是说先从 "},{"type":"codeinline","content":[{"type":"text","text":"dl_rq"}]},{"type":"text","text":" 里选择任务，然后从 "},{"type":"codeinline","content":[{"type":"text","text":"rt_rq"}]},{"type":"text","text":" 里选择任务，最后从 "},{"type":"codeinline","content":[{"type":"text","text":"csf_rq"}]},{"type":"text","text":" 里选择任务。因此，"},{"type":"text","marks":[{"type":"strong"}],"text":"实时任务总是会比普通任务优先被执行"},{"type":"text","text":"。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"heading","attrs":{"align":null,"level":4},"content":[{"type":"text","text":"调整优先级"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"如果我们启动任务的时候，没有特意去指定优先级的话，默认情况下都是普通任务，普通任务的调度类是 Fail，由 CFS 调度器来进行管理。CFS 调度器的目的是实现任务运行的公平性，也就是保障每个任务的运行的时间是差不多的。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"如果你想让某个普通任务有更多的执行时间，可以调整任务的 "},{"type":"codeinline","content":[{"type":"text","text":"nice"}]},{"type":"text","text":" 值，从而让优先级高一些的任务执行更多时间。nice 的值能设置的范围是 "},{"type":"codeinline","content":[{"type":"text","text":"-20～19"}]},{"type":"text","text":"， 值越低，表明优先级越高，因此 -20 是最高优先级，19 则是最低优先级，默认优先级是 0。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"是不是觉得 nice 值的范围很诡异？事实上，nice 值并不是表示优先级，而是表示优先级的修正数值，它与优先级（priority）的关系是这样的：priority(new) = priority(old) + nice。内核中，priority 的范围是 0~139，值越低，优先级越高，其中前面的 0~99 范围是提供给实时任务使用的，而 nice 值是映射到 100~139，这个范围是提供给普通任务用的，因此 nice 值调整的是普通任务的优先级。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https://static001.geekbang.org/infoq/75/7599c0f8ede8d3eb9f67cbf1fb177031.png","alt":null,"title":null,"style":null,"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"在前面我们提到了，权重值与 nice 值的关系的，nice 值越低，权重值就越大，计算出来的 vruntime 就会越少，由于 CFS 算法调度的时候，就会优先选择 vruntime 少的任务进行执行，所以 nice 值越低，任务的优先级就越高。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"我们可以在启动任务的时候，可以指定 nice 的值，比如将 mysqld 以 -3 优先级："}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https://static001.geekbang.org/infoq/b3/b33846f46dcceb726f2365132ec7ce0a.png","alt":null,"title":null,"style":null,"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"如果想修改已经运行中的任务的优先级，则可以使用 "},{"type":"codeinline","content":[{"type":"text","text":"renice"}]},{"type":"text","text":" 来调整 nice 值："}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https://static001.geekbang.org/infoq/d9/d950751e3443215ff8037d436bce9d01.png","alt":null,"title":null,"style":null,"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"nice 调整的是普通任务的优先级，所以不管怎么缩小 nice 值，任务永远都是普通任务，如果某些任务要求实时性比较高，那么你可以考虑改变任务的优先级以及调度策略，使得它变成实时任务，比如："}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https://static001.geekbang.org/infoq/ef/effbe4e0845544ddfba68106229b47af.png","alt":null,"title":null,"style":null,"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"horizontalrule"},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"heading","attrs":{"align":null,"level":3},"content":[{"type":"text","text":"总结"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"理解 CPU 是如何读写数据的前提，是要理解 CPU 的架构，CPU 内部的多个 Cache + 外部的内存和磁盘都就构成了金字塔的存储器结构，在这个金字塔中，越往下，存储器的容量就越大，但访问速度就会小。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"CPU 读写数据的时候，并不是按一个一个字节为单位来进行读写，而是以 CPU Line 大小为单位，CPU Line 大小一般是 64 个字节，也就意味着 CPU 读写数据的时候，每一次都是以 64 字节大小为一块进行操作。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"因此，如果我们操作的数据是数组，那么访问数组元素的时候，按内存分布的地址顺序进行访问，这样能充分利用到 Cache，程序的性能得到提升。但如果操作的数据不是数组，而是普通的变量，并在多核 CPU 的情况下，我们还需要避免 Cache Line 伪共享的问题。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"所谓的 Cache Line 伪共享问题就是，多个线程同时读写同一个 Cache Line 的不同变量时，而导致 CPU Cache 失效的现象。那么对于多个线程共享的热点数据，即经常会修改的数据，应该避免这些数据刚好在同一个 Cache Line 中，避免的方式一般有 Cache Line 大小字节对齐，以及字节填充等方法。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"系统中需要运行的多线程数一般都会大于 CPU 核心，这样就会导致线程排队等待 CPU，这可能会产生一定的延时，如果我们的任务对延时容忍度很低，则可以通过一些人为手段干预 Linux 的默认调度策略和优先级。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"horizontalrule"},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"heading","attrs":{"align":null,"level":5},"content":[{"type":"text","text":" CPU 系列好文"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"link","attrs":{"href":"https://mp.weixin.qq.com/s/TxFzIgNLettiEO4JKWgQpQ","title":""},"content":[{"type":"text","text":"CPU 执行程序的秘密，藏在了这 15 张图里"}]}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"link","attrs":{"href":"https://mp.weixin.qq.com/s/-E5jcp7tfkXjsSu2vzdeAw","title":""},"content":[{"type":"text","text":"知道硬盘很慢，但没想到比 CPU Cache 慢 10000000 倍"}]}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"link","attrs":{"href":"https://mp.weixin.qq.com/s/-uhAhBD2zGl_h19E4fNJzQ","title":""},"content":[{"type":"text","text":"如何写出让 CPU 跑得更快的代码？"}]}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"link","attrs":{"href":"https://mp.weixin.qq.com/s/PDUqwAIaUxNkbjvRfovaCg","title":""},"content":[{"type":"text","text":"10 张图打开 CPU 缓存一致性的大门"}]}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"horizontalrule"},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https://static001.geekbang.org/infoq/d8/d818ad36888ec8474c6a860ea96efc59.png","alt":null,"title":null,"style":null,"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"italic"}],"text":"哈喽，我是小林，就爱图解计算机基础，如果觉得文章对你有帮助，欢迎分享给你的朋友，也给小林点个「赞」，这对小林非常重要，谢谢你们，给各位小姐姐小哥哥们抱拳了，我们下次见！"}]}]}