Shulou(Shulou.com)06/02 Report--

This article mainly explains "what is the standard of performance tuning". The content of the explanation is simple and clear, and it is easy to learn and understand. Please follow the editor's train of thought. Let's study and learn "what is the standard of performance tuning"!

A few days ago, chatting with a classmate, he said, "our company's system has never been tuned for performance. It went online after the integration test was fine, and there were almost no performance problems after it was launched."

I didn't answer him at the time. Because there is no performance problem does not mean that the program does not have performance problems, it only means that the number of visits to the system is a little small. Do you remember that? Every time the star broke out a big melon, Weibo hung up, that is, because of the sudden increase in traffic in a short period of time, can not bear the pressure, a performance bottleneck.

Most of the performance problems are caused by too many visitors. I remember Wang Feng did a live broadcast in JD.com. The picture was so stuck that it was almost impossible to place an order and the picture could not come out. Because JD.com has never done a live broadcast before and has not encountered such a large number of visitors, it is estimated that after that event, the development of live streaming has been criticized a lot.

There are also some performance problems that break out over time, and the program will restart after running on the server for a period of time, otherwise the node memory will suddenly burst at some point. Anyway, some of our projects have encountered this embarrassment, the initial solution is to write a script, in the dead of night, secretly restart to release memory.

I think the one with the highest requirement in terms of performance is 12306, which may be scolded by people all over the country. In the past, when I was in Suzhou, no matter whether it was a holiday or not, every time I took the train back to Luoyang or from Luoyang to Suzhou, I felt that there were so many companions. Why did so many people take the train? they either went south or north, or went east or west. When it comes to Spring Festival travel, you can imagine how much pressure 12306 carries, and flash sale activity can't be compared with it at all.

If you have a partner who has worked for 12306, you can brag all your life. You are 10, 000 times better than those who worked in Taobao's Singles Day holiday (well, let me blow a wave for you first. It is said that the peak visit of 12306 is 1 billion PV, very BT).

After knowing the importance of performance tuning, let me ask my friends, when would it be better to get involved in performance tuning?

If your answer is "the sooner the better", it is obviously the wrong answer.

In my previous article, I mentioned a principle of software development, which is "Done is better than perfect" because of "perfect is never done". Performance tuning is a protracted war, and it's not a good thing to get involved early.

If you think about it, the first time the system is online is the most important, otherwise you are tired of the development progress while thinking about performance tuning, which may be a drag on the progress of system research and development. By the time your system is online, the user may have been taken away by another system, and you will never have a chance to tune your performance.

Don't always want to perfect all the functions and go online after perfection. Try and make mistakes immediately after the product has a certain prototype, according to the feedback of users, and then consider whether to add some other functions or optimization according to the needs of the market.

Let me ask my partner again, which factors will become the performance bottleneck of the system? Close your eyes, turn around and think about it.

Database: almost all systems will use the database, a large number of database read and write operations will seriously affect the performance of the system, so the database cache technology Redis is becoming more and more important. In addition, after the system has been running for a long time, the amount of data will become very large, not only in the SQL to do a lot of optimization, but also in the sub-database table.

Network: if the network speed of your home is very slow, ask the operator how much bandwidth is in your home. Or at least check to see if your network cable is 100-megabit or gigabit, even though the cable is so short that it is as short as a cat to a router. If the network bandwidth is too small, it means that data transmission is very slow, just like a sports car to the city, which may not be as fast as a bicycle.

Disk: all my desktops at home have been replaced with solid state drives, which is much faster than the previous ordinary hard drives. To say a little more, the read and write operation of the database is disk Imax O, and Redis is fast because Redis operates on memory, but the great thing about Redis is that it can not only cache, but also persist data to disk.

Memory: the read and write speed of memory is much faster than that of disk, but the space is not as large as disk. 16G memory in general home computer is already a very high configuration. As we all know, Java programs allocate memory through JVM, and the objects created are placed on heap memory, which is very fast to operate, but if there is something wrong with the code, it is easy to cause memory overflow.

The computing speed of CPU:CPU is faster than people can imagine, so AlphaGo can easily beat Ke Jie. If, I really mean if, ah, in the era of CPU or Pentium, Ke Jie could easily beat AlphaGo. If the program involves a lot of context switching, or causes frequent GC of JVM, it is easy to occupy CPU for a long time, resulting in performance problems.

In practical work, friends can also tune their performance according to the above order. At the beginning, do not blindly start on memory and CPU, which is a bit difficult, and the effect is not obvious; otherwise, it will affect the use of the whole system. Starting with database, network, and disk optimization, it's easy to see the results, and it's not easy to make mistakes.

Last question, do you guys know the performance index of the system?

1) response time

Many years ago, I did a stupid thing. The company has an idle cloud server, which is the Windows Server version. There happens to be a customer who wants to experience our system, so I didn't think so much and deployed the system to this cloud server.

How did it turn out?

The opening speed of the home page is so slow that it takes nearly a minute. Not only did the client lose his chin, but also my own.

In order to save face, I made a lot of optimizations to the back-end and front-end related to the home page. The back-end deliberately used caching technology to reduce the query of SQL statements; the front-end compressed JavaScript and CSS, reduced the number of requests, and even replaced CDN, using pictures to load lazily, but with little effect.

Finally, I thought about it, the same code, the system deployed to the Linux environment is very fast, even if the first time to open the home page needs to load resources, the response time is still too short to perceive. So the Windows Server was reinstalled as CentOS, and the effect was immediate and the response time was ridiculously short.

In that case, the response time (which refers to the time it takes for the system to respond to requests, such as when the user opens the home page) is the most obvious indicator of whether the performance of a system is good or not. For the game, the response time less than 100 milliseconds is not bad, the response time is reluctantly accepted in about 1 second, and can not be played if the response time reaches 3 seconds.

2) Throughput

Throughput (Transaction Per Second) refers to the number of transactions processed by the system per unit time, and a transaction may contain multiple requests, which reflects the pressure on the system.

It should be noted that TPS is different from QPS, which refers to the number of requests per unit time, and there is no difference between TPS and QPS when the user's operation contains only one request interface.

Throughput can be subdivided into network throughput and disk throughput. The former refers to the remaining bandwidth provided to the network application between two nodes in the network at a certain time. That is, the maximum rate that the device can accept without frame loss. The latter refers to the number of Icano requests that the system can handle per unit time, which is usually a read or write data operation request, focusing on random read and write performance.

Finally, let's make a brief summary. Performance tuning is very important, not only the user experience is good, the system is stable, but also can reflect the really excellent coding level. I'm sure all my buddies can write code that works, but as for whether it's fun or not, it needs to be studied in terms of performance.

In other words, if you jump to a company and just solve the performance bottleneck of the original system, that's great, man, you'll be reused by the company right away!

Thank you for your reading. the above is the content of "what is the standard of performance tuning". After the study of this article, I believe you have a deeper understanding of what the standard of performance tuning is. Specific use also needs to be verified by practice. Here is, the editor will push for you more related knowledge points of the article, welcome to follow!

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.