(1) The standard of performance tuning and the correct posture for tuning - if you have performance problems, go to the heapdump performance community!

Performance issues , On HeapDump Performance community !

1. When do I start to get involved in tuning

​ · The initial stage of project development , Too concerned about performance optimization , Instead, it will make us tired of wasting energy , Not only can it not bring great improvement to the system performance , It will also affect the development progress , Even get the opposite effect , Bring new problems to the system
We only need to consider the basic scenario programming performance problems at the code level ︰ Reduce disk I/О operation 、 Reduce the use of contention locks and handle external service calls carefully
​ · After development , Generally, the system will be tested . Product managers usually provide online expected data , We conduct pressure test on the reference platform provided , Through performance analysis 、 Statistical tools to count performance indicators , See if it's within expectations .
​ · After the project goes online , We need to consider the actual situation online , According to the monitoring indicators and statistical logs , To observe system performance problems , Find the problem , Fix the problem . Performance concerns core dimensions
CPU、 Memory 、 disk I/O、 The Internet 、 response time 、 throughput TPS、QPS

2. Basic steps and Strategies of performance tuning

​ Give priority to ensuring that the program is available , Then repair, troubleshooting and tuning

3. The common performance pit of strings

String、StringBuffer、StringBuilder、intern

Java9 In the program , Most strings only contain characters such as alphanumeric characters , Use Latin-1 code , One character takes up one Byte. Use Char, One Char It takes two hours Byte, It takes up twice as much memory

6 -> 7/8 Improper operation by developers may lead to error in offset statistics , Cause stack overflow . So cancel offset,count Variable

7/8 -> 9 Characters are generally alphabetic , Performance considerations use byte[]

3.1 Source code analysis StringBuffer、StringBuilder performance

StringBuffer Thread safety , But a lot of Synchronized lock .
StringBuilder It is suitable for the scenario of non competitive character resources in the concurrent environment : For example, common tool class methods .

3.2 String、StringBuffer、StringBuilder Performance pressure test

3.3 String.intern The magic of

Twitter Every time the status of a message is released , Will generate an address message . at that time Twitter Size of users , Server needs 32G To store address information . Later call String.intern, Memory becomes hundreds of megabytes

intern Method : Query whether the current string exists from the string constant pool , If it does not exist, put the current string into the constant pool

intern Method : Query whether the current string exists from the string constant pool , If it does not exist, put the current string into the constant pool

intern Method : Query whether the current string exists from the string constant pool , If it does not exist, put the current string into the constant pool

jdk6: The constant pool of strings is on the heap Perm District

JDK7、JDK8: String constant pool already from Perm Move the area to normal Java Heap Regional .Perm The area is too small is a major reason ． Often encountered OOM situation .JDK8 It has been cancelled directly Perm Area , Create a new meta region MetaSpace

3.4 String.intern Case study : Memory effect after optimization

4 Regular expression pit

use NFA Automata implement more complex regular expressions , In the process of matching, it often causes backtracking problem . A lot of backtracking takes up for a long time CPU, This leads to system performance overhead

Greedy mode (Greedy)
seeing the name of a thing one thinks of its function , It's in quantity matching , If used alone +、?、* or [min.max] Equivalent words , Regular expressions match as much as possible . For example, the previous example : text= “abbc”regex= “ab[ 1 ,3]c” In greedy mode ,NFA The automaton reads the maximum matching range , Match 3 individual b character . The match failed once , It caused a retrospective . If the match turns out to be “abbbc”, It will match

Exclusive mode (Possessive)
Exclusive mode will match more content as much as possible ; The difference is , In exclusive mode , Failure to match will end the match , There will be no backtracking problem . Previous examples , Add a... After the character “+”, Open exclusive mode .
text=“abbc”regex= “"ab[1,3]+bc The result is a mismatch . End match , There will be no backtracking problem

Previous cases are suitable for exclusive mode . Summarize regular expression skills
Use less greedy mode , Multi use exclusive mode and greedy mode will cause backtracking problems , We can use exclusive mode to avoid backtracking 2. Reduce branch selection types “(X|Y|Z)” Regular expressions that are used in this way can degrade performance , When we develop, we should try to reduce the use. If we must use , There are several ways to optimize :
First , Put the more common options in the front , So that they can be matched quickly ;
secondly , We can try to extract common patterns , for example , take “(abcd|abef)” Replace with “ab(cd|ef)"

5. java serialize : Native serialization

Java Provides a serialization mechanism , This mechanism can serialize an object into binary form ( Byte array )
Used to write to disk or output to the network , It can also read byte arrays from the network or disk , Anti serialization into objects , Use... In the program

6 NIO principle

An asynchronous non blocking I/O Model , It's also I/O The foundation of multiplexing , It has been applied to large application servers more and more , To solve the problem of high concurrency and a large number of connections 、I/O The effective way to deal with problems comes from Linux epoll Communication model

6.1 NIO Concurrent IO scene

6.2 NIO Under what circumstances can not use ?

Data communication is very frequent Don't suggest , There will be more context switches .

Suitable for use NIO Of : Long connection time , High concurrency , Communication is not very frequent

7.HashMap Blood cases and ConcurrentHashMap The correct use of

Other collection cases

HashMap The cause of the dead cycle is one : Map It's expanding 、 Multithreading Put operation

Thread safety scheme :ConcurrentHashMap

· avoid HashMap The problem of concurrent reading and writing dirty data
·JDK 1.7 ConcurrentHashmap Sectional lock , performance Synchronized Too much
·JDK1.8 Use Synchronized and CAS、Volatile、 The combination of red and black trees , Replace distributed locks