On Web Performance Optimization (1)

Pressure test

         Before talking about performance optimization , First, let's talk about stress testing . The project is before the production environment , Pressure test is required first , Simulate concurrency , Look at the throughput of the system , Inform the operation and maintenance personnel of the system throughput bottleneck , Compare with the actual business scenario , Then decide whether to continue to optimize business logic , Upgrade the hardware environment .

Concept

         Stress testing examines the current hardware and software systems Maximum load as well as bottleneck Where , It is for the system to be online processing capacity and stability Maintain within a reasonable range , Know what you know .

Wrong type

Memory leak

         For example, we write an interface , Create a large number of objects in a loop , No reuse object , When the concurrency comes up , Our objects will be constantly created , So that our memory leaks .

Concurrency and synchronization

         When we run with single thread, there is no problem , When our concurrency comes up , There will be all kinds of Thread unsafe The situation of .

Performance index of pressure test

When we run with single thread, there is no problem , When our concurrency comes up , There will be various thread unsafe situations .

Performance index of pressure test

1. response time （response time: RT）

2. HPS(hits per Second)： Hits per second ;

3. TPS(transaction per second)： Number of transactions processed by the system per second （ Business chain under a complete interface ）

4. QPS（query per second）： The number of queries processed by the system per second

Financial industry ：1000TPS-50000TPS： Not including Internet activities ,5 ten thousand

The insurance industry ：100TPS-100000TPS： Not including Internet activities ,10 ten thousand

Manufacturing industry ：10TPS-5000TPS,5K

Internet e-commerce ：10000TPS-1000000TPS,100w

Medium sized Internet websites ：1000-5wTPS

Small Internet sites ：500-1wTPS

Maximum response time ： We tested 100w Request , The response time of one request is 90s;

Looking from the outside , Performance testing focuses on the following three indicators

1. throughput ： The number of requests and tasks that the system can process per second （qps and tps）

2. response time ： The time it takes for a service to process a request or a task

3. Error rate ： The proportion of requests with wrong results in a batch of requests

  Pressure test tools -JMeter

The throughput of stress test is relatively small , We can adjust JVM Memory

windows Port occupation error and its solution ：

influence web The main factors of performance

database

Applications

middleware （tomcat、nginx、gateway gateway ）

The Internet IO( The current network speed is slow , Or the server is under pressure , Or the current bandwidth is only 1M--10000 request , Each request 1KB The data of =10MB, It will be affected by bandwidth )

operating system （ Different systems , The kernel is different , Different ways of handling ） Other aspects .

Classification of factors affecting performance

First of all, consider that your application belongs to CPU intensive （ A lot of calculation and sorting ） still IO intensive （ The Internet IO And disk read and write IO）;

Watch the surveillance ：cpu and io And the occupation of memory and network traffic ;

If it is CPU intensive ： Upgrade the server hardware | Increase the number of servers

If it is IO intensive ： Replace the SSD + Memory module + Using caching technology + Improve the transmission efficiency of network card , Simply add cpu Can't solve the problem .

Performance monitoring and analysis

java Memory model analysis

Program counter ：Program Counter Register

· The record is Being implemented Of Virtual machine bytecode Of Instruction address ;

· This memory area is the only one in JAVA Nothing is specified in the virtual machine specification OutOfMemeoryError Region .

Virtual machine stack ：VM Stack

· Describe the JAVA Method Memory model of execution , Each method creates a Stack frame , be used for Store local variable table , The stack of operands , Dynamic links , Method interface Etc .

· The local variable table stores various basic data types known to the compiler , Object reference .

· If the stack depth requested by thread is not enough, it will report StackOverFlowError abnormal .

· If the dynamic expansion capacity of the stack is not enough, it will be reported that OutOfMemeoryError abnormal .

· The virtual machine stack is thread isolated , That is, each thread has its own independent virtual machine stack .

Native Method Stack ：Native Stack

The local method stack is similar to the virtual machine stack , It's just that the local method stack uses local methods .

Pile up ：Heap

Almost all object instances and arrays allocate memory on the heap ; The heap is the main area managed by the garbage collector , Also known as “GC” Pile up , It's also the place that we consider most about optimization .

About heap memory reclamation process

         All object instances and arrays are allocated on the heap . The heap is the main area of the garbage collector , Also known as GC Pile up , It is also where we optimize most .JVM Use C Written language , Can open up space , It's troublesome to free memory . be based on C Repackage , Write an execution engine , Go and translate it java Code .

Heaps can be subdivided into

1. The new generation

   1. Eden Space ： The newly assigned object arrives eden District , First, check whether the memory is enough , What if it's enough , Just allocate memory directly ; If not enough , It's going to take place once GC,（ Do it once. young GC, Also called MinorGC, Mainly clean up the Cenozoic space ,eden There is... In it 10 Objects , An object is in use , rest 9 Objects are no longer used , Just put this 9 Kick it out , For this object, we will put it in the survivor area . If minorGC I can't put it down after that , We think this object is a big object , Let's try to put it in the old age --> The old age is when our new generation cannot deal with it . If the old age still can't let go , Then we will have a fullGC, The holocaust , Put the data of all regions of the new generation and the old generation , If it's useless , All kicked out , After kicking out , If you can't put it down , Then an error is reported, and the memory overflows .oom-outofmemory out of memory ） If the survival time of the object in the survivor area exceeds the threshold （ A dozen times of small massacres are still alive ）, Just move him to the elderly area .

   2. from suivivor Space 【s0】

   3. to sruvivor Space 【s1】

2. Old age

3. Forever / Meta space

java8 The old generation , suffer jvm management ,java8 Later, it will be changed to meta space , Use physical memory directly . therefore , By default , The size of the meta space is limited only by local memory .

utilize java Memory monitoring tools for tuning

jconsole And jvisualvm

jvisualvm Can do

Monitor memory leaks , Tracking garbage collection , Memory at execution time ,cpu analysis , Thread analysis, etc .

State of thread

function ： Running ;

Sleep ：sleep

wait for ：wait

Resident ： Idle threads in the thread pool

monitor ： Blocked threads , Waiting for lock

  Monitor all docker Container of cpu Usage rate docker stats

As can be seen from the figure above ,nginx More wasteful cpu;

JMeter Simulate and analyze concurrent requests

 

The more middleware , The greater the performance loss , Most of them are lost in the network interaction between middleware .

1. First, let the throughput of the middleware itself be improved

2. Improve the transmission efficiency between middleware （ Ethernet cable , network card , Transfer protocol ）

Business ：

db：MYSQL Optimize

The rendering speed of the template （thymeleaf Cache on ,CPU, Memory ）

Nginx Separation of motion and stillness

Summarize the ways to improve performance ：

1. Turn on thymeleaf The cache of ( Close when developing , Turn on when running )

2. Close the server log （ Not recommended for use ）

3. nginx To separate the movement from the rest

4. Database indexing

5. Reduce database interaction （ Request all the data for reuse java To deal with ）【 Optimize the business 】

6. Improve memory ,jvm The new generation is enlarged, etc

7. Cache

Next ：

introduce redis Cache problems and solutions