当前位置:网站首页>Aren't you curious about how the CPU performs tasks?

Aren't you curious about how the CPU performs tasks?

2020-11-09 10:50:09 InfoQ

{"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":" Preface "}]},{"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":" Do you know the following questions ?"}]},{"type":"bulletedlist","content":[{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":" With memory , Why do we still need CPU Cache?"}]}]},{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"CPU How to read and write data ?"}]}]},{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":" How to make CPU It's faster to read data ?"}]}]},{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"CPU How pseudo sharing happens ? And how to avoid ?"}]}]},{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"CPU How to schedule tasks ? If your task is highly responsive , You want it to always be scheduled first , What should I do ?"}]}]},{"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":" This article , Let's answer these questions ."}]},{"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":" Text "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"heading","attrs":{"align":null,"level":3},"content":[{"type":"text","text":"CPU How to read and write data ?"}]},{"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":" Let's get to know CPU The architecture of , Only understand CPU Of framework , To better understand CPU How to read and write data , For the modern CPU The architecture of is as follows :"}]},{"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":" You can see , One CPU There are usually multiple CPU The core , Like the one above 1 Number and 2 Number CPU The core , And each CPU The core has its own L1 Cache and L2 Cache, and L1 Cache Usually divided into dCache( Data caching ) and iCache( Instruction cache ),L3 Cache It's shared by multiple cores , This is it. CPU A typical cache hierarchy ."}]},{"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":" All of the above are CPU Inside Cache, If you look outside , There will also be memory and hard disk , These storage devices together constitute the pyramid storage hierarchy . As shown in the figure below :"}]},{"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":" As you can see from the picture above , From the top down , The larger the capacity of the storage device , And the slower the access speed . As for the access delay of each storage device , You can see the chart below :"}]},{"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":" You can see , CPU visit L1 Cache Faster than accessing memory 100 times , That's why CPU There will be L1~L3 Cache Why , The purpose is to put Cache As CPU Cache layer between and memory , To reduce the frequency of memory access ."}]},{"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 Read data from memory to Cache When , It's not a byte by byte read , It's block by block , This piece of data is called CPU Line( Cache line ), therefore "},{"type":"text","marks":[{"type":"strong"}],"text":"CPU Line yes CPU Read data from memory to Cache The unit of "},{"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":" as for CPU Line size , stay Linux The system can view , You can see my server's L1 Cache Line Size is 64 byte , That means "},{"type":"text","marks":[{"type":"strong"}],"text":"L1 Cache The size of the data loaded at a time is 64 byte "},{"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":" So loading arrays , CPU It will load multiple consecutive data in the array to Cache in , Therefore, we should access the elements in the order of physical memory address distribution , When accessing array elements like this ,Cache The hit rate will be high , So you can reduce the frequency of reading data from memory , This can improve the performance of the program ."}]},{"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":" however , We don't use arrays , But when you use separate variables , Will have a Cache The problem of pseudo sharing ,Cache Pseudo sharing is a performance killer , We should avoid it ."}]},{"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":" Next , Take a look at Cache What is pseudo sharing ? And how to avoid this problem ?"}]},{"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":" Now suppose you have a dual core CPU, these two items. CPU The core runs two different threads in parallel , They read two different data from memory at the same time , They are of type "},{"type":"codeinline","content":[{"type":"text","text":"long"}]},{"type":"text","text":" The variable of A and B, The addresses of these two data are in physical memory "},{"type":"text","marks":[{"type":"strong"}],"text":" continuity "},{"type":"text","text":" Of , If Cahce Line Its size is 64 byte , And variable A stay Cahce Line At the beginning of , So these two data are located in "},{"type":"text","marks":[{"type":"strong"}],"text":" The same Cache Line in "},{"type":"text","text":", Again because CPU Line yes CPU Read data from memory to Cache The unit of , So these two data will be read into two at the same time CPU Each in the core Cache in ."}]},{"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":" Let's think about a problem , If these two threads with different cores modify different data , such as 1 Number CPU The core thread only modified Variable A, or 2 Number CPU The thread of the core thread only modifies the variables B, What will happen ?"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"heading","attrs":{"align":null,"level":4},"content":[{"type":"text","text":" Analyze the problem of pseudo sharing "}]},{"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":" Now let's combine to ensure the consistency of multi-core cache MESI agreement , To illustrate the whole process , If you don't know MESI agreement , You can read my article 「"},{"type":"link","attrs":{"href":"https://mp.weixin.qq.com/s/PDUqwAIaUxNkbjvRfovaCg","title":""},"content":[{"type":"text","text":"10 A picture opens CPU The door to cache consistency "}]},{"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. First variable A and B Not yet Cache Inside , hypothesis 1 Core No. 1 binds threads A,2 Core No. 1 binds threads B, Threads A Only read and write variables A, Threads B Only read and write variables 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 Core read variables A, because CPU Read data from memory to Cache Its unit is Cache Line, It's just the variable A and Variable B The data belongs to the same Cache Line, therefore A and B The data will be loaded into Cache, And Cache Line Marked as 「 Monopoly 」 state ."}]},{"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. next ,2 The core starts to read variables from memory B, It's also about reading Cache Line The size of the data to Cache in , this Cache Line The data in also contains variables A and Variable B, here 1 Number and 2 No. 1 core Cache Line The status changes to 「 share 」 state ."}]},{"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 Core No. 2 needs to modify variables A, Found this Cache Line The state of is 「 share 」 state , So we need to send a message to 2 Number one core , notice 2 No. 1 core handle Cache Corresponding Cache Line Marked as 「 Has lapsed 」 state , then 1 The core of No Cache Line The state becomes 「 The modified 」 state , And modify the variable 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. after ,2 Core No. 2 needs to modify variables B, here 2 No. 1 core Cache Corresponding Cache Line It's a failed state , In addition, due to 1 No. 1 core Cache There's the same data , And the state is 「 The modified 」 state , So we need to 1 No. 1 core Cache Corresponding Cache Line Write back to memory , then 2 And then read it from memory Cache Line The size of the data to Cache in , Finally, put the variable B Modify to 2 No. 1 core Cache in , And mark the status as 「 The modified 」 state ."}]},{"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":" therefore , You can see if 1 Number and 2 Number CPU In this way, the core continuously alternates the variables A and B, Will repeat ④ and ⑤ These two steps ,Cache It doesn't work as a cache , Although variable A and B There is no relationship between them , But because it belongs to one at the same time Cache Line , This Cache Line After any data in has been modified , Will affect each other , So that ④ and ⑤ These two steps ."}]},{"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":" therefore , This is because multiple threads read and write the same Cache Line Different variables of , And lead to CPU Cache The phenomenon of failure is called *"},{"type":"text","marks":[{"type":"italic"}],"text":" False sharing ("},{"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":" How to avoid pseudo sharing "}]},{"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":" therefore , For hot data shared by multiple threads , Data that is often modified , It should be avoided that these data happen to be in the same Cache Line in , Otherwise, there will be the problem of pseudo sharing ."}]},{"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":" Next , Let's see how to avoid the problem of pseudo sharing in the actual project ."}]},{"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":" stay Linux There is... In the kernel "},{"type":"codeinline","content":[{"type":"text","text":"__cacheline_aligned_in_smp"}]},{"type":"text","text":" Macro definition , Is used to solve the problem of pseudo sharing ."}]},{"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":" From the above macro definition , We can see :"}]},{"type":"bulletedlist","content":[{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":" If it's multi-core (MP) In the system , The macro definition is "},{"type":"codeinline","content":[{"type":"text","text":"__cacheline_aligned"}]},{"type":"text","text":", That is to say Cache Line Size ;"}]}]},{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":" And if you're in a single core system , The macro definition is empty ;"}]}]}]},{"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":" therefore , For the same Cache Line Shared data in , If there is serious competition between multiple cores , In order to prevent the occurrence of pseudo sharing , You can use the macro definition above to make the variable in Cache Line It's aligned inside ."}]},{"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":" for instance , There's this structure :"}]},{"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":" Two member variables in a structure a and b Continuous in physical memory address , So they might be in the same place Cache Line in , Here's the picture :"}]},{"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":" therefore , In order to prevent the above mentioned Cache Pseudo sharing problem , We can use the macro definition described above , take b The address of is set to Cache Line Align address , as follows :"}]},{"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":" such a and b Variables will not be in the same Cache Line It's in , Here's the picture :"}]},{"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":" therefore , avoid Cache Pseudo sharing is actually the idea of exchanging space for time , Waste a part of Cache Space , In exchange for performance improvement ."}]},{"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":" Let's look at an application level circumvention scheme , There is one Java Concurrent framework Disruptor Use 「 Byte padding + Inherit 」 The way , To avoid the problem of pseudo sharing ."}]},{"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 There is one of them. RingBuffer Classes are often used by multiple threads , The code is as follows :"}]},{"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":" You might think RingBufferPad In class 7 individual long The name of the type is strange , But in fact , They don't seem to work , But it plays a crucial role in improving the performance ."}]},{"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":" We all know ,CPU Cache The unit of reading data from memory is CPU Line, commonly 64 position CPU Of CPU Line Its size is 64 Bytes , One long The type of data is 8 Bytes , therefore CPU It will load later 8 individual long Data of type ."}]},{"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":" according to JVM Object inheritance relationship in the parent class members and child class members , Memory addresses are arranged sequentially , therefore RingBufferPad Medium 7 individual long Type data as Cache Line "},{"type":"text","marks":[{"type":"strong"}],"text":" Prepopulation "},{"type":"text","text":", and RingBuffer Medium 7 individual long Type data is used as Cache Line "},{"type":"text","marks":[{"type":"strong"}],"text":" Post fill "},{"type":"text","text":", this 14 individual long Variables have no practical use , And they don't read and write them ."}]},{"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":" in addition ,RingBufferFelds The variables defined in it are "},{"type":"codeinline","content":[{"type":"text","text":"final"}]},{"type":"text","text":" Embellished , It means that it will not be modified after the first load , also "},{"type":"text","marks":[{"type":"strong"}],"text":" because 「 Before and after 」 Each is filled with 7 One that can't be read or written long Type variable , So no matter how you load Cache Line, This whole Cache Line There is no data in the database that will be updated , So as long as the data is read and accessed frequently , Naturally, no data is swapped out Cache The possibility of , Therefore, there will be no pseudo sharing problem "},{"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 How to select the thread ?"}]},{"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":" To understand the CPU After reading the data , Let's see CPU What is the basis for selecting the thread to be executed ."}]},{"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":" stay Linux The kernel , Both processes and threads use "},{"type":"codeinline","content":[{"type":"text","text":"tark_struct"}]},{"type":"text","text":" Structure represents , The difference is that threads are tark"},{"type":"text","marks":[{"type":"italic"}],"text":"struct Some resources in the structure share the resources created by the process , For example, memory address space 、 Code segment 、 File descriptors, etc , therefore Linux Threads in are also called lightweight processes , Because of the thread tark"},{"type":"text","text":"struct Compared to the process of tark_struct Carrying Less resources , So in order to 「 light 」 name . "}]},{"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":" Generally speaking , Processes that do not create threads , There is only a single execution flow , It's called the main thread . If you want a process to handle more things , You can create multiple threads to process , But anyway , They all correspond to the kernel "},{"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":" therefore ,Linux The scheduler in the kernel , The object of scheduling is "},{"type":"codeinline","content":[{"type":"text","text":"tark_struct"}]},{"type":"text","text":", Next we'll call this data structure "},{"type":"text","marks":[{"type":"strong"}],"text":" Mission "},{"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":" stay Linux In the system , According to the priority of the task and the response requirements , There are two main types , The smaller the value of priority , The higher the priority :"}]},{"type":"bulletedlist","content":[{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":" Real time tasks , It requires high response time for the system , That is to do real-time tasks as quickly as possible , The priority is "},{"type":"codeinline","content":[{"type":"text","text":"0~99"}]},{"type":"text","text":" Within the scope of the real-time task ;"}]}]},{"type":"listitem","content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":" General mission , Response time is not very demanding , The priority is "},{"type":"codeinline","content":[{"type":"text","text":"100~139"}]},{"type":"text","text":" It's all at the normal task level ;"}]}]}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"heading","attrs":{"align":null,"level":4},"content":[{"type":"text","text":" Scheduling class "}]},{"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":" Because of the priority of tasks ,Linux In order to ensure that high priority tasks can be executed as early as possible , So it is divided into these scheduling classes , Here's the picture :"}]},{"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 and Realtime These two scheduling classes , They're all applied to real-time tasks , These two scheduling classes have three Scheduling Strategies in total , Their functions are as follows : "}]},{"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":": Is in accordance with the deadline On schedule , The nearest to the current point in time deadline Will be scheduled first ;"}]}]},{"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":": For tasks of the same priority , According to the principle of first come, first serve , But higher priority tasks , Can preempt low priority tasks , That is to say, those with high priority can 「 Jump the queue 」;"}]}]},{"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":": For tasks of the same priority , Running in turn , Each task has a certain time slice , Tasks that run out of timeslices are put at the end of the queue , To ensure the fairness of the same priority tasks , But high priority tasks can still preempt lower priority tasks ;"}]}]}]},{"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":" and Fair Scheduling class is applied to ordinary tasks , It's all by CFS Scheduler managed , There are two scheduling strategies :"}]},{"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":": The scheduling strategy used by common tasks ;"}]}]},{"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":": The scheduling strategy of background tasks , Don't interact with the terminal , So it doesn't affect other tasks that need interaction , You can lower its priority appropriately ."}]}]}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"heading","attrs":{"align":null,"level":4},"content":[{"type":"text","text":" Completely fair scheduling "}]},{"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":" The basic tasks we meet in our daily life are , For normal tasks , Fairness is the most important thing , stay Linux Inside , It is based on CFS Scheduling algorithm of , That is to say *"},{"type":"text","marks":[{"type":"italic"}],"text":" Completely fair scheduling ("},{"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":" The idea of this algorithm is to let each task be assigned CPU Time is the same , So it schedules a virtual runtime for each task vruntime, If a task is running , The longer it runs , Of the mission vruntime The bigger it will be , And tasks that are not running ,vruntime It won't change ."}]},{"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":" that ,"},{"type":"text","marks":[{"type":"strong"}],"text":" stay CFS When scheduling algorithms , Preference will be given vruntime Less tasks "},{"type":"text","text":", To ensure the fairness of each task ."}]},{"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":" This is like comparison. , Let you divide a bucket of milk tea equally into 10 A cup of milk tea, a cup of , Which cup of milk tea do you see less , Just pour more ; Which one is more , Don't fall first , After several rounds of operation , Although there is no guarantee that every cup of milk tea is exactly the same , But at least it's Fair ."}]},{"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":" Of course , The example mentioned above does not consider the issue of priority , Although it's a normal task , However, there is a priority distinction between ordinary tasks , So in computing the virtual runtime vruntime Also consider the common task "},{"type":"text","marks":[{"type":"strong"}],"text":" Weight value "},{"type":"text","text":", Note that the weight value is not a priority value , There will be one in the kernel nice Conversion table of level and weight values ,nice The lower the level, the greater the weight value , as for nice What is the value , We'll talk about later ."}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":" So here's the formula :"}]},{"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":" You can leave it alone NICE"},{"type":"text","marks":[{"type":"italic"}],"text":"0"},{"type":"text","text":"LOAD What is it? , You think it's a constant , So in 「 The same actual run time 」 in , High weight tasks vruntime Than low weight tasks vruntime "},{"type":"text","marks":[{"type":"strong"}],"text":" Less "},{"type":"text","text":", You may wonder why there is less ? You remember CFS Dispatch , It will give priority to vruntime Fewer tasks to schedule , So high weight tasks will be scheduled first , Therefore, the actual running time of high weight will naturally be more ."}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"heading","attrs":{"align":null,"level":4},"content":[{"type":"text","text":"CPU Run the queue "}]},{"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":" A system usually runs a lot of tasks , The number of multitasking is far more than CPU The core number , So it's time to "},{"type":"text","marks":[{"type":"strong"}],"text":" line up "},{"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":" in fact , Every CPU All have their own *"},{"type":"text","marks":[{"type":"italic"}],"text":" Run the queue ("},{"type":"text","text":"Run Queue, rq"},{"type":"text","marks":[{"type":"italic"}],"text":")"},{"type":"text","text":"*, Used here to describe CPU All processes running on , Its queue contains three running queues ,Deadline Run the queue dl"},{"type":"text","marks":[{"type":"italic"}],"text":"rq、 Real time task run queue rt"},{"type":"text","text":"rq and CFS Run the queue csf"},{"type":"text","marks":[{"type":"italic"}],"text":"rq, among csf"},{"type":"text","text":"rq It's described in terms of red and black trees , Press vruntime Size to sort , The left most leaf node , It's the task that will be scheduled next time ."}]},{"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":" These scheduling classes have priority , The priorities are as follows :Deadline > Realtime > Fair, It means Linux Choose when the next task is performed , It will be selected in this order of priority , That is to say, first from "},{"type":"codeinline","content":[{"type":"text","text":"dl_rq"}]},{"type":"text","text":" Choose the task in , And then from "},{"type":"codeinline","content":[{"type":"text","text":"rt_rq"}]},{"type":"text","text":" Choose the task in , Finally from the "},{"type":"codeinline","content":[{"type":"text","text":"csf_rq"}]},{"type":"text","text":" Choose the task in . therefore ,"},{"type":"text","marks":[{"type":"strong"}],"text":" Real time tasks always take precedence over normal tasks "},{"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":" Adjust priorities "}]},{"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":" If we start the mission , If you don't specify priorities , By default, they are normal tasks , The scheduling class for common tasks is Fail, from CFS Scheduler to manage .CFS The purpose of scheduler is to realize the fairness of task running , That is to say, the running time of each task is almost the same ."}]},{"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":" If you want to have more time for a normal task , You can adjust the task "},{"type":"codeinline","content":[{"type":"text","text":"nice"}]},{"type":"text","text":" value , This allows higher priority tasks to execute more time .nice The range that can be set is "},{"type":"codeinline","content":[{"type":"text","text":"-20~19"}]},{"type":"text","text":", The lower the value , Indicates that the higher the priority , therefore -20 It's the highest priority ,19 It's the lowest priority , The default priority is 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":" Do you think nice The range of values is weird ? in fact ,nice Value does not indicate priority , It's a modified value for priority , It's related to priorities (priority) The relationship is like this :priority(new) = priority(old) + nice. The kernel ,priority The range is 0~139, The lower the value , The higher the priority , In front of them 0~99 Scope is provided for real-time tasks , and nice Values are mapped to 100~139, This scope is provided for common tasks , therefore nice Value adjusts the priority of normal tasks ."}]},{"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":" We mentioned earlier , Value and weight nice Value of the relationship between ,nice The lower the value , The greater the weight value , Calculated vruntime The less , because CFS When scheduling algorithms , You'll give priority to vruntime Fewer tasks to carry out , therefore nice The lower the value , The higher the priority of the task ."}]},{"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":" We can start the mission , You can specify nice Value , For example, will mysqld With -3 priority :"}]},{"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":" If you want to change the priority of a task that is already running , You can use "},{"type":"codeinline","content":[{"type":"text","text":"renice"}]},{"type":"text","text":" To adjust nice value :"}]},{"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 Adjust the priority of common tasks , So no matter how small nice value , Tasks are always ordinary tasks , If some tasks require high real-time performance , Then you can consider changing the priority of tasks and scheduling strategies , Make it a real-time task , such as :"}]},{"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":" summary "}]},{"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":" understand CPU Is the premise of how to read and write data , To understand CPU The architecture of ,CPU Internal multiple Cache + External memory and disk constitute the pyramid of memory structure , In this pyramid , Lower and lower , The larger the memory capacity is , But the access speed will be small ."}]},{"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 Reading and writing data , It's not a byte by byte basis , But rather CPU Line Size is in ,CPU Line The size is usually 64 Bytes , That means CPU Reading and writing data , Every time it's with 64 The byte size operates as a block ."}]},{"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":" therefore , If the data we're manipulating is an array , So when accessing array elements , Access according to the address order of memory distribution , This will make full use of Cache, The performance of the program has been improved . But if the data of the operation is not an array , It's a normal variable , And in multicore CPU Under the circumstances , We also need to avoid it Cache Line The problem of pseudo sharing ."}]},{"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":" So-called Cache Line The problem of pseudo sharing is , Multiple threads read and write the same Cache Line Different variables of , And lead to CPU Cache Failure phenomenon . So for the hot data shared by multiple threads , Data that is often modified , It should be avoided that these data happen to be in the same Cache Line in , The ways to avoid it are Cache Line Size byte alignment , And byte padding ."}]},{"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":" The number of multithreads that need to run in the system is generally greater than CPU The core , This will cause the thread to queue and wait CPU, This may cause some delay , If our task has a low tolerance for delay , It can be intervened by some artificial means Linux The default scheduling policy and priority of ."}]},{"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 Series of good articles "}]},{"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 The secret of executing the program , It's hidden here 15 Zhang Tuli "}]}]},{"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":" Know the hard drive is slow , But I didn't expect to CPU Cache slow 10000000 times "}]}]},{"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":" How to write let CPU Faster code ?"}]}]},{"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 A picture opens CPU The door to cache consistency "}]}]},{"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":" hello , I'm Xiao Lin , I love to illustrate the basics of computer , If you think the article will help you , Welcome to share with your friends , And give Xiao Lin some 「 Fabulous 」, This is very important for Kobayashi , thank you all , Let's give you a big hug , I'll see you next time !"}]}]}

版权声明
本文为[InfoQ]所创,转载请带上原文链接,感谢