How to plan the data center and build an optimized operation?

Some ideas and experiences of our company, today's main content is about how to locate the management of data center infrastructure. Traditionally, data centers are usually isolated from IT. In the era of cloud computing, we will think that all services in the data center should be treated as an infrastructure, just like Microsoft's container, which already has the center of IT technology. At this level, it can be started, and our IT can only get our network and our server. So IT means that we consider the environment of IT architecture and the whole environment as a unified object and consider their mutual management, which is different from our traditional data center.

From the life cycle of a data center, from the beginning of design, we will find that this is a data management after continuous operation. In the design stage, we can see that as an infrastructure, IT department is not specific to several servers. At this time, we need to consider the computing power from the perspective of capacity, that is, how much computing power the data center can support. Therefore, the data center must have such a capacity, which cannot be estimated at 60. The whole IT estimation process is at least 1-3 years. After the design is completed, we will implement it. Implementation is actually a deployment process. After the deployment is completed, the next step is an operation. The operation process needs to solve a big problem, whether the actual operation is consistent with our original design. If it is inconsistent, we will make adjustments, including more adjustments and rectification. Assuming that the actual situation is consistent with the expected situation, there is not much work to be done at the operational level. Now the IT situation is becoming more and more complicated. From the IT point of view, the input cost is greater than our original budget. The last stage is the optimization stage. We are constantly adjusting and the demand is constantly changing. We need optimization. The result of optimization is the design of the next data center model, which becomes a cycle. In terms of data center infrastructure, it has five physical requirements, including attack, refrigeration, space, security and wiring. Our standards for purchasing IT equipment must also be unified. If these standards are mainly embodied in the five physical requirements we have just refined, we should pay attention to these five aspects in the design stage. But in today's data center, users often think more about power supply and heat dissipation, because there is a demand for computing power, and 90% of the electricity consumption of all servers will be discharged, so heat dissipation is a very important factor. At the management level, there are three very important places. First of all, in order to ensure the controllability of IT, we are generally divided into three levels: bottom infrastructure, top IT and top business applications. For example, the IT layer provides services similar to the business layer. If the business level is not available, it means that the necessity of IT existence will be challenged. Therefore, ensuring availability is the first priority. If the first priority is met, we can reduce the availability. For the operation of data center, we can divide it into several stages. The first stage is deployment, which is the process of supporting IT operations. The second stage is that the IT business system is online, and the business has started to operate. We can use all kinds of OA. The third stage is the monitoring system. When problems are found, corresponding actions can be taken to ensure the normal operation of the business system. Fourth, under the condition of ensuring business availability, some technologies or cloud technologies can be used to realize green optimization and reduce operating costs. In the data center, infrastructure currently faces five major challenges. The first point is usability, which is also a very important place for our existence. The last two points are the elements of our continued existence in the cloud era, including adaptability and expansibility in the life cycle. Finally, manageability and maintainability/service orientation are interdependent.

Back to the initial design stage, it doesn't seem to matter much, because in the design stage of our company, it basically doesn't exist, just a general requirement. When the data center is running, we can show the data operation through this diagram to know whether the aircraft seats are overheated and whether the power supply is sufficient. Because our assumptions are different from the actual situation, we need to manage these data. We try our best to narrow the huge gap between reality and design. Of course, it is impossible to narrow the gap. Otherwise, our business will not develop. When there is no way to narrow the gap, we must add a module to expand it. This is also the definition of data center modularity. At present, we can narrow the gap from two aspects. One is to establish hard-support infrastructure, including adjustable modular infrastructure, for example, modular UPS. Modular power supply and modular air conditioning can realize the design of infrastructure. In addition, the power distribution and capacity management system can help us grasp the digestion trend of infrastructure. We can understand the future development trend of IT, including new technologies such as cloud computing, so that we can adopt better technologies in the data center and make this data center have stronger vitality.

Second, by optimizing the use of infrastructure, do some comprehensive, detailed and accurate understanding, by planning the use of infrastructure, by extending the use of data centers, real-time detection of infrastructure, eliminate abnormal situations, so that infrastructure can be used well.

Suppose we are now maintaining a good data center. Actually, digging is very important. We often say that our network is unmanageable. At present, most of the infrastructure is manageable. When we face a new data center, or when we adjust the data center, the first step is to dig out some functions that were originally invested, such as the previous refrigeration UPS. After digging out the manageable functions, it is a very important step to integrate into the new platform, which involves limitations, mainly in data compatibility and refined management. It can be said that for a management interface, the strength of management determines the depth of management. So in the design stage, we decide our own management depth and manageable luminosity. At this level, we decide what equipment we need to buy. If we invest a lot of money to buy a multifunctional device, we find that we can't manage it. In the management of infrastructure, we seldom talk about the word management, but mostly use the word monitoring. In China's vocabulary, monitoring has two meanings. Inspection is to detect, collect and collect a large amount of data, and control is to make some feedback, adjust the running state, and how to decide the strategy according to the collected running data. For most users, most of them are detecting, which is an automatic or instinctive monitoring means. If this is a cloud data center, we may adjust it manually, but we should judge it according to the collected data, and then decide how to control it according to the business. This control process is also automatic, so that the response speed can be fast enough to meet the requirements of cloud computing. It also means that when testing, we decide the amount of information collected by the device. In the whole process, we need to consume the test data we collected, plus our own knowledge base of these data, and finally form a control result, which is also the core of management.