MasterofProject

The resource allocation mechanism of cloud Tencent Quest

Label Tencent cloud
1543 people read Comment(0) Collection Report

One This is the QCon Shanghai 2013 interview room, we are now very happy to go to the head of Tencent cloud Chen Lei.

Chen Lei:Very pleased to have the opportunity to have a dialogue with friends in this InfoQ, thank you for your attention.

   

Two Tencent recently made a very large online activities, and then we are very concerned about. So a lot of people are asking a very key problem, that is, Tencent cloud, Tencent itself in the end is how to use? Tencent in the end what business are running on the Tencent cloud? What kind of situation is the situation of the business operation? We can refer to Tencent's business model to bring their own similar business to Tencent cloud?

Chen Lei:I just shared the Qzone of the InfoQ, which is in the use of Tencent cloud, including its internal virtualization technology, which uses the network to accelerate technology and storage services. Qzone the use of storage services, with us to our customers, Tencent cloud customer service is exactly the same.

   

Three Is the same technology stack is it?

Chen Lei:Is the same technology stack, and is the same service. For example, Qzone CDB, with our customers with the CDB is the same, but different clusters, including the software version is exactly the same, the deployment of the program is also exactly the same.

   

Four What about Qzone?

Chen Lei:In addition to Qzone, QQ, micro channel, our core business, both in the use of these services. From the whole of the technical architecture, we are the same.

   

Five So you can rest assured that with QQ tied together. We see Tencent cloud in the choice of computing resources, unlike other clouds, only a few fixed options. Tencent cloud is free to choose the configuration, and I believe that if you can configure freely, it will produce a special combination, the management of the cloud will be a big problem. I want to know, Tencent cloud in the management of such a variety of virtual models, what kind of things to do, these virtual machine is how to allocate? Is there a number of resources for the super sale? If there is a resource to seize the situation, the last is how to schedule?

Chen Lei:First of all, there is no overbooking mechanism, we have absolutely no oversold. Indeed, it is like a sensible, if the scheduling scheme is not reasonable, there may be an empty. We are a resource hot spots for the priority of the algorithm, for example, in my entire Site inside, when the memory resource becomes hot, then I will give priority to meet the needs of the user's memory. I will go to see which resource is the most valuable in the present situation. Because from the point of view of the whole room, it can just be the memory, CPU, hard disk resources are all sold out, this is not possible.

   

Six It's very hard, right.

Chen Lei:Well, because the user's needs are all together, with the capacity of your resources, it is not likely to be exactly the same. So you have to go first to ensure that the most scarce resources, to ensure the efficiency of its use, which is the most important. In addition, in our entire distribution algorithm, it is also good to consider the filling and isolation of the hole. Because our business is to follow the law, the user's choice may not be all sorts of strange things. We see the user's choice is divided into two major categories, a class of CPU and memory will be relatively high, you will choose a high configuration of virtual machine; the other may be on the CPU and memory requirements are not so high, they will choose a relatively small, will be very long tail. So we put the long tail and high demand models to mix. We see the result is that the use of resources in these different business itself has focused, in fact, a complementary situation. In the sale of resources, we are not considered oversold.

   

Seven That I have a problem, it is difficult to say that a user in its virtual machine is not up, it is difficult to determine whether it is in the end what is dense, such as computing intensive or memory intensive business, then you are how to judge the business to be placed. Because he wants to buy a resource, you give him a resource, and then to see what kind of it is, right?

Chen Lei:In fact, it is not difficult to judge. Because you have a history of the purchase and use of history. For a new customer, you generally know it's the industry. For example, games, the use of resources is a certain feature of the game. Although every game developers, there will be some slight differences, but the game, especially now we are such a web game or mobile games, he is very standard for the requirements of resources. And memory is a very important resource for them. If it is a site, it may be another form of allocation of resources, so you still have a lot of methods to do a certain prediction, of course, this prediction is not wrong? It may be wrong.

   

Eight Is there a correction, is also to give it heat transfer, or some other difference?

Chen Lei:We don't do this. But we'll do it. In addition to the use of our CEE users, then we deal with more flexible. I can give it two different instance, for example, in different devices, he may use the initial amount of small, I just give him two different Container. In the use of it, we will be synchronized to expand the two Container, when the expansion to a certain extent, I will find that this machine is empty, the machine is full. Then I will give him some resources in the machine is empty, the machine with relatively less to him some resources. In addition we can implement a very fast heat transfer, Container heat transfer rate is very high.

   

Nine Have you realized the heat transfer of Container?

Chen Lei:Well.

InfoQ:This is a relatively new technology, it should be said that the industry should be more than the lead, right?

Chen Lei:Well.

   

Ten I have a question is, Tencent in the data center disaster recovery, there is no any particularly good technology to guarantee? Because a lot of people, including in the use of some foreign cloud users. In fact, they are want oneself to care about whether the data centre is down, so I need to put my things in several of the world data center deployment, explicit to allow the user to do this thing. That Tencent cloud in this regard is what kind of technology to ensure that?

Chen Lei:Simply saying that the user explicitly to choose, may still not be perfect to meet the actual needs of the user. Users want to take the initiative to pick, I want to deploy my equipment in Shanghai and Tianjin two places, often a purpose. For example, it may choose to go to Shanghai, to the south of the user distribution, or it may be distributed to Tianjin to the north of the user to carry out the distribution, we also have this kind of developers is specifically to pick the. However, the most sophisticated technology is in fact the storage of synchronization. If you are a logical Server, or Web server, the simple distribution is in fact enough, but the most important problem is how to store the data is very good to do remote synchronization.

   

Eleven This is the cross data center of this?

Chen Lei:We are now using the internal synchronization center, is currently open to our partners to use gray scale. Our future standard version of the CDB is to support multi data center automatic synchronization, so the problem of data synchronization, you can help users solve these complex technical problems, solve the worries. So just go to pick up the right point, and then deploy its logic and Web servers, including its software.

   

Twelve That just mentioned Container, we said that we are now more concerned about a problem, that is, the security of the Container, the isolation of the end? Including foreign reports, the standard Container is a known vulnerability, may appear to escape, if someone do this thing, will not affect other users?

Chen Lei:Container's isolation, we are doing a number of kernel changes. But we did not do one hundred percent of isolation. At present, the biggest problem is that there are some system parameters of the configuration is global. Then changes to the system parameters will affect the other Container. We take the way is to prohibit users from modifying. In addition to the Container itself is a technology that can be Share, when some services beyond the scope of its use of resources, it is to provide a number of mechanisms for Share. Will have an impact on other business and technology of the container itself is said, if the resource is not in short supply, you can eat too much. When resources are scarce, Container will be strict limits on your quota, so we think it's a little problem for this technology.

   

Thirteen Tencent will be the Container related security mechanisms, the contribution to the community to change the mechanism of the community? Because I have seen the recent changes in the kernel, there are a lot of people on the Container submitted a lot of patch, Tencent has such an idea?

Chen Lei:In fact, the amount of changes we submitted to the kernel, in the whole of China is concerned, it may be more than HUAWEI. We are about to be second, so we have already submitted a lot of patch to the kernel, including some of the Container we have already submitted. Of course, not all of our changes are submitted, there are some of us want to be better able to verify, give yourself more time to ensure that we submitted to the quality of the community.

   

Fifteen Users voted yes, right?

Chen Lei:Help us choose, right. We do the CE (engagement customer), found in fact, we are the majority of users, the current is still dominated by Git, SVN we will certainly support.

InfoQ:Just a matter of time, isn't it?

Chen Lei:Right. Just a matter of time.

   

Sixteen There is also a problem in the use of the cloud, we will use the cloud database, not to build a. So Tencent in providing cloud database, is how to ensure the isolation between the user? In the specific implementation, you are a relatively large database of changes? Or do some other things, based on other and existing software, so the performance and security is how to guarantee?

Chen Lei:This problem is particularly simple, we give different users of different instance, that is, we will not let the two users go to Share with a MySQL database. From the safety and the point of view, this is the most effective and the most effective. These databases are set up on the cloud, they share the underlying resources. But every user to see the database, is only its own, so there will not be a logical or security issues.

   

Seventeen This is quite interesting. Because if you do so, you don't need to maintain your own code, right?

Chen Lei:Unwanted

InfoQ:Or is it easy to be a solution?

Chen Lei:Yes

   

Eighteen I on the site, look at the entire structure of the Tencent cloud, as if there is no block storage mode, I do not know this is a kind of consideration, is there such a plan?

Chen Lei:In fact, already in the gray, we already have a certain amount of users in the use of our cloud disk. We give customers a product of the time, there is a relatively long gray cycle, because in the introduction of new products we are more cautious, do not want to launch a less sophisticated products to customers.

   

Ninteen Do you have a standard, what can you offer to the user? Can you talk about this standard specifically?

Chen Lei:We have a very specific process. The first is in the internal business use, then we will go to see the failure rate as well as the type of failure. Actually is a kind of quality of the QA monitoring, the control. When we are in the internal use, no longer be able to find a comparison of the impact of the Bug, we will consider opening up to the outside. The process of opening up is a very cautious gray process, we will be with those who we are very good, particularly willing to forgive our customers, to say that we have a very mature product. But it has been used in the QQ, the inside of the Qzone has been used, you can not see if you have a try? During the trial of the gray, we will give a very deep discount, sometimes it is completely free of charge. During this period we will go to see the user's use of the scene will not be a problem. We have the necessary cycle, such as gray level must be more than three months. When we get to the key nodes of the three month, we will Bug in this period of time to find the Review, will go to the Review users to use the scene, to assess whether there is enough representation. If it is not sufficiently representative, we may go to add some users, and then carry out a wider range of gray. Like cloud disk, we have been gray for almost half a year.

   

Twenty That still can be expected to have a long time to have this expectation?

Chen Lei:Should be in the near future, it will be fully open to release.

   

21 After all, it's a very important function. We talked about block storage, and it is very related to the block is a mirror image services, Image services. Many people may be more concerned about, there are some foreign countries have a number of mature models, that is, I posted on the cloud system my own mirror, and then you can buy. So Tencent will not have a mirror image of the service or the market to consider it?

Chen Lei:Mirror image services we are also in the gray process, has been to some game developers to use. Because the game developers we know that it is very high demand for open service, especially some of the game developers with Windows technology, the demand for mirror technology will be particularly high. Because of his software installation process is the need to intervene, so we first give them a mirror image of the service. Now in the gray process, it is expected to be open in the near future.

   

22 We also know that a complete cloud, not only has a cloud provider, but also around the cloud providers have some of the surrounding ecosystem, there are a lot of people do value-added services, it is likely to be some of the message, or SMS service, such a small service. So Tencent has such an idea, you can let the third party service access in, and give them a good growth in space?

Chen Lei:Sure, we did a few things. First of all, we will provide a very rich API, you can go to operate our cloud products. And our API is compatible with Amazon's API.

InfoQ:Really? How compatible?

Chen Lei:We are not so satisfied, but we should eventually be one hundred percent compatible.

InfoQ:One hundred percent compatible, right?

Chen Lei:Our goal is. We want to be able to run on the Amazon all the tools you can run directly on the Tencent cloud.

   

23 Amazon is a huge system, you have to do the same thing?

Chen Lei:We may be based on some of the main services, we will focus on some of the points, but we provide the service of the API will be compatible with the amazon.

InfoQ:That's too good. Heard so many very exciting news, I am very optimistic about the future development of Tencent cloud, and then very much thank Chen Zong today to the QCon site to accept our interview.

Top
0
Step
0
Guess what you're looking for.
View comments
* above the user's comments only represent their personal views, do not represent the views or position of the CSDN website
    Personal data
    • Visit15336 times
    • Integral:163
    • Grade
    • Rank:Miles away
    • Original0
    • Reprint:9
    • Translation:0
    • Comment:2
    Article Archive
    Latest comments