Chapter 643 Cost control is very important

Time synchronization in data centers is very important.

For example, in the payment system, there is 100 yuan in the account, and I bought virtual products such as point cards in the north. If you can't synchronize in time, you will find that the 100 yuan in Nanfang's account has not been spent, and you can buy some cards again.

These days, netizens take advantage of technical loopholes to buy Q coins for free.

Domestic is fine.

According to Cheng Binghao, the synchronization system developed by Ziweixing is guaranteed to be launched within 3 months. Although the country is very large, it is landlocked and connected by optical fiber, so the technical difficulty is not high.

But Facebook is doing it differently. People are playing global synchronization, which is world-class difficulty.

In the eastern hemisphere and the western hemisphere, there is a delay in making a phone call. God knows what method they will use to synchronize the database in time.

Then, Tang Binchen briefly talked about the other 7 technical projects.

The flow control system means that during the peak period, the server and broadband pressure of the website are too high, which will cause data transmission congestion and data loss. In order to avoid this situation, it is necessary to control the flow of some users with good computer performance and bandwidth usage, and distribute resources to more users.

Well, useful for friends.com.

The distributed cache system is to allow the system to perform cache reading in parallel to improve efficiency and save page response time. Behind the scenes, different cache forms should be selected according to the amount of tasks to save space and further improve the reading speed.

This project is very technically difficult.

Well, it's useful for Moments.

Disaster recovery system, no matter how good the server is, no matter how advanced the operating system is, server failures will definitely occur. If a small number of servers fail to respond, other servers will be responsible. When a large number of servers are unresponsive, all the load must be borne by a small number of servers, which can be very dangerous and result in data loss and corruption. Therefore, it is necessary to set a threshold value. When the threshold value is reached, the page will no longer fail to respond, so as to ensure the stability of the data center.

Speaking of this, Zhou Buqi is very familiar with it.

This kind of situation often occurs in the "Double Eleven" and the rush to buy train tickets during the Spring Festival. This situation also occurs when Xiaonei.com sells popular products.

Zhou Buqi said with a smile: "We also have this system, which was imitated from Wangsu, and it existed two years ago."

Wang Xingdao: "It's not surprising, Facebook has just started its global deployment, and their server operations are not under much pressure, and they should have never encountered such a situation before."

Tang Binchen went on to say: "The dispatching and monitoring system means that when servers process some tasks in parallel, different servers must monitor each other. Once a problem occurs on one side, the other side will respond immediately and take over the unfinished tasks."

Zhou Bu's head is getting bigger, "Parallel? Distributed algorithm?"

"right."

"Facebook is also engaged in distributed algorithm research?"

"This is a big trend." Tang Binchen knew what Boss Zhou meant, and smiled to comfort him, "But everyone is doing distributed, more data processing."

Zhou Buqi asked: "Do you know the process? How is Facebook's progress in this regard?"

"Is it a distributed algorithm?"

"right."

"It is precisely because of the progress of this algorithm that they can carry out research on caching systems and monitoring systems."

Zhou Buqi's complexion is not very good-looking, "How long have they been doing this?"

Tang Binchen said: "Three or four months."

"Depend on!"

Zhou Buqi collapsed a little, deeply feeling the technical gap.

Ziweixing is working on a distributed database. First of all, it needs to do a distributed algorithm. It has been almost a year, and the second wave of 100 million R\u0026D funds will be poured into it again.

But it is far from the application.

Looking at Facebook again, after three or four months of work, the distributed algorithm can be easily applied.

Once reborn, Zhou Buqi can make earlier layouts on some application layers, but the technical level... is really a bit weak, and early layouts can't change the actual conditions of backward technology. He is a layman, and he can't even participate if he wants to.

Fortunately, the country is developing very fast, and more and more Chinese in Silicon Valley are bringing back advanced thinking and technological concepts on their way back to China. The speed of technological progress in China far exceeds that of Silicon Valley.

Wang Xing saw that Boss Zhou was a little depressed, so he smiled and said with relief: "The distributed database is a big project, but the distributed cache system and the scheduling monitoring system are small projects, which are different. Our distributed algorithm needs to be refined, and Facebook only needs Simple applications are fine. From the perspective of distributed algorithms alone, we will be able to launch the application in May, and the functions will be more powerful and stable.”

Zhou Buqi's eyes lit up, "It will be available in May?"

Wang Xingdao: "Yes, it has already been made. Yang Zhenkun is paying attention to the details... The main reason is that the investment is too large. This is the first time that the product of the R\u0026D center has appeared. He wants to be foolproof, so there will inevitably be delays in time." .”

Zhou Buqi asked: "When our distributed algorithm is finished, we can also do the research and development of the cache system and monitoring system?"

Tang Binchen and Wang Xing looked at each other, and gave a rather vague answer, "Almost."

No one knows what's going on in the R\u0026D center.

All I know is that the money was spent, and I didn't see any results.

Yang Zhenkun monopolized the power, Xu Liangjie went to inspect it twice, but it was useless to quarrel twice. Boss Zhou has great trust in Dr. Yang.

The next few system projects are all about improving the page response time. Through technical means, users can open the webpage faster. At the same time, it is necessary to reduce the CPU usage on the web side, reduce server requirements, and so on.

And here, there is a huge difference in the concept of using the database.

As we all know, in the era of information explosion, the traditional database model can no longer keep up and must be improved. Ziweixing chose the distributed database under the distributed server, and the multiple single-row databases under the distributed server architecture adopted by Facebook.

It is to create a database for the information flow, photos, logs, shared links, reposts and other content on the website. Use the form of multiple databases to deal with explosive data and improve reading efficiency.

In order to facilitate Boss Zhou's understanding, Tang Binchen gave an example——

If you compare the database to a pot, and compare the data to beans, there are four types: red beans, mung beans, black beans, and white beans. Reading information is picking beans.

The traditional stand-alone database mode is to pick up beans by one person, which is very inefficient.

Ziweixing's mode is to put all the beans in one pot, and four people pick them up together, which improves the efficiency.

Facebook's model is to divide all beans into four pots, one red bean pot, one mung bean pot, one black bean pot and one white bean pot. Then four people correspond to four pots, and the efficiency is greatly improved.

Zhou Buqi frowned, "It sounds like theirs is more high-end?"

Tang Binchen said with a smile: "It's an analogy. We are gangsters, and they are single gangsters. Random gangsters are safer. If one person dies, the remaining three people will not be affected. It is very stable. They are gangsters, and they are single-machine structures. The database, so the person who takes the red beans can only take the red beans, and the person who takes the mung beans can only take the mung beans. If the person who takes the red beans dies, the red beans can’t be taken out.”

Zhou Buqi said: "So their model is not as stable as ours? So they need to spend more on hardware conditions?"

Tang Binchen said: "Yes, from the perspective of efficiency, they are higher. From the perspective of cost and stability, it is our advantage."

Zhou Buqi understood.

No wonder those giants in Silicon Valley do not engage in distributed databases, but are actively seeking technical upgrades to traditional servers such as oracle and mysql.

They rely on the global market and are so rich that they don't care about this small amount of money. Stability can be compensated with money, nothing more than building more servers and more standby databases.

If a person is asked to pick up red beans, if the person dies, the red beans cannot be picked up, and the system will collapse. But if 100 people are dedicated to picking red beans, if a group dies, the red beans can still be taken out.

Wang Xingdao: "Silicon Valley does not engage in distributed databases because the market structure has already been set, and database giants like Oracle have dominated for many years. Engineers learn to use databases like Oracle when they are studying, and let them change later. , and can’t be changed. The only way to upgrade the use of traditional databases is to develop a distribution system for beans and a variety of storage systems for different beans through technical means.”

Zhou Buqi understood what he meant, "So we need to start training from scratch? Give the university money, set up a special laboratory, and set up courses and research topics for our 'Feitian' database?"

Wang Xing nodded, "It is very necessary."

Tang Binchen said: "Distribution must be the future trend. Compared with the single-row structure, the parallel structure has huge advantages. Or the example just now, as long as we also engage in data distribution and different storage systems, and then divide the beans into different In the basin, isn’t it more efficient when grabbing? It also has a mixed inspection function and strong stability.”

Zhou Bu's head is getting bigger.

Brain full of red beans.

However, a new perspective appeared in his mind.

The domestic database market is blank, which may not be a good thing. Silicon Valley is dominated by giants, so it is very difficult to innovate and innovate, and it is a bit rigid.

The country is a blank sheet of paper, and it happens to be able to splash ink.

It's like Musk making rockets, screwing with space agencies all over the world, and then... through cost control, amazing results have been achieved.

If the Starlink project uses the traditional rocket launch mode, the cost will be too high, and it is simply unaffordable. Musk's rocket can be recycled and reused, giving this plan a feasible idea.

Technology is important, and cost control is also very important!

There will only be more and more data in the future.

Traditional one-row databases backed by high-end, expensive servers certainly won't work.

There must be a cheaper alternative.

This is the opportunity of distributed databases.

After dismissing Tang Binchen and Wang Xing, Zhou Buqi leaned weakly on the executive chair, feeling that his brain cells were almost exhausted. It was really embarrassing for him to participate in such a complicated technical discussion as a layman.

What are the examples of red beans, mung beans, and white beans? It's a mess!

"Ya Xian!"

"coming."

"Call and ask where Yang Mi is. I'll go to her in a while."

"What are you doing?"

"It's nothing, just eat red beans and chat."