MySQL combat optimization expert 05 production experience: how to plan the database machine configuration in the real production environment?

1、 When you understand the principle of database architecture , You should know your database Planning

We used 4 This article gives you an overall analysis MySQL How databases work , I believe many friends have learned a lot about the overall architecture principle of the database I have a definite understanding of , After all, before that , Maybe most people are right about MySQL The understanding of database still stays in execution SQL Degree of statement .

When we have a preliminary understanding of the database architecture design principle , Then the first thing you should know , Is that we usually work , How to plan database . Because I think many people are mainly responsible for some things without much concurrency , A system with dozens or hundreds of users , Then I won't go at all Pay attention to the planning of the database .

Yes, a lot Java For Engineers , Or find one by yourself linux The machine is equipped with one MySQL, Then connect your system and start using it directly , want Otherwise, let DBA Or the operation and maintenance engineer helps him find a machine and install one MySQL perhaps Oracle, Then you can use it directly .

But in our column , We hope to teach you more professional database experience , Including the overall architecture principle of the database , There is also how to plan Database in production environment , Including when you have a production Library , The thing to do is to design the pressure measurement scheme , Including the pressure test of your database , Including your Database deployment visual monitoring system , wait .

When you do this series of things , Then you should develop your Java System , To operate your database , Realize various business functions and logic .

2、 What is the general configuration of the production database ？

Now let's look at the first question , Suppose you need to deploy a database in the production environment , At this point, you first need a machine to deploy the database . Then I What we should consider is , Deploy a database of production environment , What kind of machine configuration is generally required ？

Next, we will tell you some of our experience values , Directly tell you what kind of machine configuration is deployed MySQL database , Roughly suitable for how high concurrent visits Ask quantity

When you know this experience value , In the future, when you are in charge of system development , When applying for the database , You will know what kind of database is needed in the production environment The machine is equipped with , Roughly how many concurrent accesses can be resisted .

First of all, let's be clear , If the system you are responsible for is one with little concurrent access , A system with dozens or hundreds of users , So in fact, you choose What kind of machine to deploy the database , The impact is not great , Even if you use our personal laptop to deploy one MySQL database , Its It can also support the operation of low concurrency systems .

Because that kind of system may send a wave of requests to the database every few minutes , And there may be hundreds of tables in the database 、 Thousands or tens of thousands of data , Small amount of data , The concurrency is very small , The operating frequency is very low , The number of users is very small , The concurrency is very small , It's just that the business logic of the system may be very complex . For this kind of system Database machine selection , It's not in our consideration .

We mainly focus on Internet Systems with a certain amount of concurrency , It may generate hundreds of data per second , Thousands per second , Even tens of thousands of concurrent requests per second , For such scenarios , What kind of machine should we choose to deploy the database , In order to better resist the pressure of our system .

3、 ordinary Java How much concurrency can the application system resist when deployed on the machine ？

Generally speaking , According to our experience ,Java When deploying the application system, the commonly used machine configuration is 2 nucleus 4G and 4 nucleus 8G More , Database department The machine configuration often used in the deployment is at least 8 nucleus 16G above , Normal in 16 nucleus 32G

Based on our production experience of a large number of highly concurrent online systems , commonly Java The application system is deployed in 4 nucleus 8G On the machine , Every second 500 about The number of concurrent visits , It's almost appropriate , Of course, this is not necessarily . Because you have to think about it , Suppose you spend 1s It can be done , So you one This machine may only be able to process 100 A request , But if you only spend per request 100ms You can finish , Then your machine may be able to Processing hundreds of requests .

So how many requests per second can a machine resist , It is often related to how long it takes you to process each request , But generally speaking , Based on our extensive experience For observation ,4 nucleus 8G The machine deployment is ordinary Java Application system , It can resist hundreds of concurrent accesses per second , From onetwohundred requests per second to sevenoreight hundred requests per second seek , It's all possible , The key is to see how long it takes you to process each request .

4、 High concurrency scenarios , What kind of machine should the database use ？

For databases , We just said , Generally, the recommended database is at least selected 8 nucleus 16G With the machine , Even 16 nucleus 32G Our machine is more suitable some .

Because you have to consider a problem , For our Java Application system , The main time-consuming thing is Java Network communication between system and database . Yes Java The system itself for , If you just run some common business logic inside the system , Complete some business logic in your own memory , This performance is extremely high .

For you Java Every request received by the system , The most time-consuming thing is to send network requests to the database , Wait for the database to execute some SQL sentence , Return the result to you .

So we often say that you have one Java The system is under great pressure , The load is high , But in fact, you should understand one thing , You this Java In fact, the main pressure and complexity of the system are Focus on the one you rely on MySQL On Database ！

Because you perform a lot of additions, deletions, modifications and checks SQL At the time of statement ,MySQL The database needs a lot of memory and disk files IO operation , So databases are often The one with the highest load ！ We have this question before 4 In this article , adopt MySQL Analysis of the principle of database architecture , All have been explained .

And your Java Generally, the system does not require you to read and write a large number of local files directly, which is time-consuming IO Let's do it ？ Is it right? , I think I did Java A friend of development will Will want to understand this truth .

So often for a database , They all choose 8 nucleus 16G As the starting machine , It's best to choose 16 nucleus 32G Our machine is more suitable , Because the database needs Execute a large number of disks IO operation , Each of his requests is more time-consuming , So the configuration of the machine naturally needs to be higher .

Then from our previous experience , commonly 8 nucleus 16G The machines are deployed MySQL database , It's OK to resist one or two thousand concurrent requests per second , But if you The concurrency of is higher , Suppose there are thousands of concurrent requests per second , Then maybe the database will be a little dangerous , Because the database CPU、 disk 、IO、 Memory load Will be very high , If the pressure on the database is too high, it will go down .

about 16 nucleus 32G The machines are deployed MySQL Database , Two or three thousand per second , Even three or four thousand concurrent requests are OK , But if you reach every Tens of thousands of requests per second , So the database CPU、 disk 、IO、 The load of memory will soar to a high level in an instant , The database may not be able to withstand downtime .

So this is for databases , We generally recommend the configuration of the selected machine , And how much concurrent requests he can resist .

In addition, for databases , If you can , It's better to use SSD Solid state drive instead of ordinary mechanical hard disk , Because the biggest complexity of the database lies in a large number of disk IO, He needs a lot of disk files to read and write , So if you can use SSD Solid state disk , Then your database can resist more concurrent requests per second .

5、 Today's question

Today I want to leave you with a small question to think about ： Suppose you developed Java The system is deployed on a 4 nucleus 8G On the machine , So let's assume this Java The system processes a request Beg very, very fast , Each request only needs 0.01ms You can handle it , So you think this machine is deployed Java System , It can resist thousands of concurrency per second Please ？ Can you resist tens of thousands of concurrent requests per second ？